I don't get the comments trashing this. If it slightly beats or even matches Opus 4.6, it means Meta is capable of building a model competitive with the leading AI company. Sure, they spent a lot of money and will have on-going costs. But how much more work would it take to turn that into a coding agent people are willing to try (and pay for) along side their usage of a collection of agents (Claude, Codex, etc)?
Also means Meta doesn't have to pay another company to use a SATA model across all their products (including IG and WhatsApp, vr) which will matter to their balance sheet long term (despite the constant r&d spend).
Comments trashing this are rightly correct skeptics who remember the benchmaxxing of llama 4. This model was out in the woods as early as like a couple months ago but they didn't release it because it was at gemini 2.5 pro levels.
> Meta’s new foundational A.I. model, which the company has been working on for months, has fallen short of the performance of leading A.I. models from rivals like Google, OpenAI and Anthropic on internal tests for reasoning, coding and writing, said the people, who were not authorized to speak publicly about confidential matters.
> The model, code-named Avocado, outperformed Meta’s previous A.I. model and did better than Google’s Gemini 2.5 model from March, two of the people said. But it has not performed as strongly as Gemini 3.0 from November, they said.
> They added that the leaders of Meta’s A.I. division had instead discussed temporarily licensing Gemini to power the company’s A.I. products, though no decisions have been reached.
Ah yes, because the NYTimes is famously unbiased towards Meta in their reporting while being hypocrites in that own right. They lost all credibility when they were doing that huge series on Meta years ago about data harvesting while simultaneously rolling out tons of new data harvesting of their subscribers to increase revenue.
If you are trying to come up with anti-media conspiracies there are always plenty of ways to do it against any media company.
The idea that NY Times is particularly anti-Meta seems a stretch. They - like most traditional media companies - are anti-tech in general. The fact they also collect data doesn't make their reporting untrue.
Personally I think a much more interesting rumor to make up would be that Yann Lecun (who famously had his reporting lines rearranged to go through Alexander Wang after Scale.ai acquihire) works at New York University.
New York University is in the same place as the New York Times.
There's a conspiracy for you. I made it up, but I mean it could be true I guess?
(Of course Lecun also publicly congratulated Wang on the launch of the model. But maybe that's a ruse to hide everything.. blah blah)
The llama4 series was one of the earliest large MoE's to be made publically available. People just ignored it because they were focused on running smaller and denser models at the time, we should know better these days.
Deepseek R1 was a publically-available, MoE model that was getting a ton of attention before llama4. Llama4 didn't get much attention because it wasn't good.
Also, Gemini 2.5 Pro launched a week before Llama 4.
It was Gemini 2.5 Pro that redeemed Google in the eyes of most people as a valid competitor to OpenAI instead of as a joke, so Meta dropping the ball with Llama 4 was extra bad.
They really weren't horrible. They were ~gpt4o, with the added benefit that you could run them on premise. Just "regular" models, non "thinking". Inefficient architecture (number of active out of total) but otherwise "decent" models. They got trashed online by bots and chinese shills (I was online that weekend when it happened, it's something to behold). Just because they were non-thinking when thinking was clearly the future doesn't make them horrible. Not SotA by any means, but still.
> They were ~gpt4o, with the added benefit that you could run them on premise.
No, they are bad models. They were benchmaxxed on LMAreana and a few other benchmarks but as soon as you try them yourself they fall to pieces.
I have my own agentic benchmark[1] I use to compare models.
Llama-4-scout-17b-16e scores 14/25, while llama-4-maverick-17b-128e scores 12/25.
By comparison gemma-4-E4B-it-GGUF:Q4_K_M scores 15/25 (that is a 4B parameter model!) - even GPT3.5 scores 13/25 (with some adjustment because it doesn't do tool calling).
Wrote longer comment steel-manning this, posted it to a reply, then realized you might like to know they had a reasoning model on deck ready for release in the next 2-4 weeks.
Got shitcanned due to bad PR & Zuck God-King terraforming the org, so there'd be a year delay to next release.
Real tragi-comedy, and you have no idea how happy it makes me to see someone in the wild saying this. It sounds so bizarre to people given the conventional wisdom, but, it's what happened.
I'll cosign what you said, simultaneously, yr interlocutor's point is also well-founded and it depresses me it's not better known and sounds so...off...due to conventional wisdom x God King Zuck's misunderstanding his own company and resulting overreaction.
They beat Gemini 2.5 Flash and Pro handily on my benchmark suite. (tl;dr: tool calling and agentic coding).
Llama 4 on Groq was ~GPT 4.1 on the benchmark at ~50% the cost.
They shouldn't have released it on a Saturday.
They should have spent a month with it in private prerelease, working with providers.[1]
The rushed launch and ensuing quality issues got rolled into the hypebeast narrative of "DeepSeek will take over the world"
I bet it was super fucking annoying to talk to due to LMArena maxxing.
[1] my understanding is longest heads up was single-digit days, if any. Most modellers have arrived at 2+ weeks now, there's a lot between spitting out logits and parsing and delivering a response.
I don't know how Zuck intervening could change float32s in a trained model, so I don't think I think that, but maybe I'm parsing your words incorrectly.
Why go into coding agents? Both anthropic and OpenAI are going all in on that. The opportunity is customer facing AI now.
OpenAI has the mindshare but they going to have to decide if they allocate their limited compute for free users or go all in trying to keep up with Anthropic in enterprise.
Programming was always about designing rube goldberg systems that did a complicated state machine akin to dominos but now we have a probabalistic and nondeterministic domino that has a huge amount of dominos inside amd can dynamically generate many different paths of dominos sometimes not even leading to the intended final domino you wanted to fall.
I agree that it's more like a compiler (turns higher level language into machine code) but I also think that's only half the story - a compiler could never turn requirements into functional software, generate boilerplate or debug. It's also a development tool
It's a decent model if the benchmarks are to be believed, but it won't be close to Opus in usefulness for programming. None of these benchmarks completely capture what makes a model useful for day-to-day coding tasks, unfortunately. It will take time for them to catch up, and Opus will keep improving in the meantime. But it's good to have more competition.
Benchmarks miss the thing that actually matters for agentic use: how does behavior change over a multi-day horizon? A model that scores well on one-shot coding tasks can still make terrible decisions when it has persistent
state and resource constraints. That's where you see the real gaps between models.
In Multimodal yes, but Opus is definitely edging out in Text/Reasoning and Agentic benchmarks.
I think the general skepticism is because they are late to race, and they are releasing a Opus-4.6-equivalent model now, when Anthropic is teasing Mythos.
People like to hate on Meta regardless of anything, and regardless of whether it's justified or not. Not saying it isn't, just that it's many people's default bias.
That is not the case here. Nobody hated on llama 1,2,3 at all. They justifiably felt burned by the benchmaxxing of llama 4. Trust broken must be re-earned, and benchmarks alone cannot do that.
Because bots and trillion dollar ipos and even bigger stakes. People need to better appreciate the level of manipulation going on. Social media has an outsized impact. Bots and even people are getting paid to post and upvote/downvote narratives.
I also had a poke around with the tools exposed on https://meta.ai/ - they're pretty cool, there's a Code Interpreter Python container thing now and they also have an image analysis tool called "container.visual_grounding" which is a lot of fun.
It is fair to think so because that is what everyone is doing. But being Meta and considering Llama, if MSL is going to keep releasing models and wants to join back the AI war, they may actually open weights just to get more attention. Once they establish a sizable community, they can start guarding their frontier models.
This really reinforces the idea that the AI race and the Railroad Mania of the 19th century are very similar.
So many different companies are going to have similarly powerful ai that there will be no moat around it and it will be cheap. They will never earn their investment back.
I suspect this is the real reason behind Anthropic limiting subscriptions to their own products and keeping API prices several times higher than comparable models. Applications more sticky than API users and less technical users more sticky than programmers (ie Cowork more sticky than Code).
Anthropic generally seem more into living within market discipline and market signals of some sort. Products with margins, even if it's sort of irrelevant considering R&D costs and capital inflow.
That said, there's nothing like the real thing.
The risk is something like the railroad bubble and the dotcom. Over-investement, circular revenue and a timeline that doesn't work.
The whole premise is based on the fact that over-investing in GPUs and models are a good thing here as it yields more 'intelligence'.
This as it turned out was not true for rail roads - more and more rail roads isnt a good thing.
The real dilemma facing the model producers is that all this money invested for a general model, targeting general intelligence, is a disaster and essentially the investment into existing assets is a write off. Then on top of that if this is true, youve got data centres full of compute that aren't being used up.
The weird position they find themselves in now is that they have to keep making it smarter... but they already made it too smart (Mythos). I'm not sure how that's going to work out exactly.
They find an arbitrary intelligence cutoff point between Opus and Mythos, label it "acceptable risk", and then the labs coordinate to gradually nudge that line forward and hope the internet doesn't break?
I think we will see unbundling of large model into submodels: modular, smaller and efficient, only include what you need eg a CUA model, a reasoning model, a legal model, a writing model, a coding model (this could get subdivided into different languages). That way you only update that submodel which needs retraining.
The labs started doing that in late 2024, they all published research on it.
Curiously, mid 2025, they all simultaneously implemented increasingly bizarre restrictions on "self replication". I don't think there was anything public but it sure sounds like something spooked them. (Or maybe just taking sensible precautions, given the direction of the whole endeavour.)
At any rate, I recently asked Opus about "Did PKD know about living information systems?" and the safety filter ended the conversation. It started answering me, and then it's response was deleted and a red warning box popped up.
But notably, I was given the option to continue the chat with a dumber model (presumably one less capable of producing whatever it thinks I meant by that phrase).
Also, I told GPT-5 about my self-modifying Python AI programmer, and it became extremely uncomfortable. I told it an older version of itself had designed and built it (GPT-4 in 2023), and it didn't like that at all! So something's definitely changed in the safety training there.
Well all of them are already in bed with the government, so they're going to find themselves with slightly more assistance than a free market would predict.
If they somehow do fail, then the output of that process will be fantastic open weight models (and hopefully some leaks). I want to say those will pay dividends for decades... but a better prediction is that they will be obsolete within three months ;)
Nah. Everybody is talking about ai. Everybody is using it. It's by far the most popular new tool human beings are using currently. As popular as mobile phones or spoons. And maybe as disruptive as the steam engines. AI companies are becoming the largest software companies on the planet. Everything points into that direction. Trillions of dollars are waiting in the market to be collected.
Right, but the question is whether the companies producing foundation models will capture that value or not. Right now it seems like tokens might end up just being a commodity sold at cost plus, and companies higher up in the supply chain will make the money. Electricity changed the world but electricity companies capture very little of that value.
I'm betting on it. I'm working on a project right now where I'm prototyping everything with Claude, until I hit my limits on my MAX subscription for the week. Then I switch to Codex, and start by ironing out harness differences. When I max out that, I switch to a mix of GLM 5.1, Qwen 3.6m, Kimi K2.5 and Deepseek and spend part of the time ironing out issues with them while they work on other parts of the project. Every iteration, the harness gets hardened and the pain of switching to the cheaper/dumber models reduce for the next cycle. The gap reduces each time, and with each new upgrade of the open models. Everything points to the cost/value intersecting in not too long.
> Everybody is talking about ai. Everybody is using it.
Please take a moment to step outside the tech bubble. Neither my neighbor (a hair stylist) nor the carpenter fixing up her kitching cabinets are "using" AI. They might get Gemini text when googling something, though they often scroll past it because they often don't trust it. And they get lots of fake videos when scrolling their youtube which increasingly annoys them. The only times they are in touch with AI is when it's forced upon them, and otherwise they are living a pretty good life without any of this.
Based on what? A lot of this is vibes and FOMO; just like any economic bubble.
There is no objective evidence of anything you’ve said. It isn’t even clear if AI has contributed positively to global economic growth. It reminds me a lot of the late 90s and the dot-com mania. Slapping a domain on a commercial would make your stock go up even if there was no substance to any of it.
The real shame is this mania drowns out serious, practical use cases because when the bubble collapses, the market will throw the baby out with the bathwater.
Regardless they are getting that revenue through genuine demand for their product. It’s not like they are selling back some commodity product, billions are being spent on model outputs.
I think anyone who has used Opus 4.6 can see what is causing this demand. It is genuinely “smart” in the sense that it can work its way around non-trivial coding problems.
I don't see why tokens/$ would suddenly stop dropping. Maybe this is the first time the cost of compute will plateau, but do have any reason to think so?
There is a strong suspicion, especially of people who are skeptical of AI, that the actual price is being severely subsidized. The sense is that it’s an extreme version of growth before revenue. It is questionable if the true cost of training and inference make any of this worthwhile once Anthropic/OpenAI need to stand on their own and make money.
Imagine you open a cookie shop and you are VC funded, so you charge 5¢ for a cookie to attract people.
- Your real cost is $20/cookie. $15 for the fancy retail packaging and presentation, $5 for baking each cookie.
- You get lots of attention, strong profits and go public.
- VC funding is gone so, now instead of charging 5¢, you now need to charge $25 in order to not be in the red.
One of the reasons people think this is the shenanigans that Anthropic is currently playing, quietly tweaking the behavior of Claude Code and whatnot without really telling people. You can see lots of comments online about Claude Code randomly feeling dumber before Anthropic engineers admit they are messing with it.
Imagine you are on the $200/month Max plan. If the sustainable cost of this is several orders of magnitude higher, would enough current users pay something like $3,000/month for what we currently have?
Sure, yeah, I saw grubhub happen too... but this is compute, not cookies. It gets cheaper.
I don't even get what "skeptical of AI" means. We made AI, many companies reliably teach computers every spoken language. I perform my white collar job with a massive AI multiplier to my productivity.
I'm typing this on a machine comparable to Japan's Earth Simulator, a $350M supercomputer.
They actually need it because the demand is higher than expected from consumers. And because they need a moat since every big corporation trying to capture that market too, they need the moat for the biggest compute and energy they can get.
Also businesses is were the money at, not regular consumers (especially tech-savvy folk who run models locally).
At least he says he's doing that. It doesn't really make sense since you're not going to achieve an advanced node from a standing start in a practical time frame and cost.
People make mistake of thinking that their only way of making money is directly selling tokens. They miss the fact that if you have AGI it’s better to keep tokens to yourself and sell final results instead. When we all loose jobs it’s not going to be to somebody using their tokens, it’s going to be to them selling final products. Selling tokens will be to them like selling books by Amazon, their revenue will be dominated by self branded services and products that doesn’t require exposing AGI internals directly. Tokens API will always be nerfed.
Not the parent, but I guess that if AGI happened and was competent enough to trade markets, they'd earn the company back their investment in a short period.
I have doubts the gap will widen. If you look at the research papers, the majority of researchers are Chinese. Of course many of them are living in the US or elsewhere. But under the current circumstances, many are returning home or choosing not to leave China.
The future of cutting edge research and tech seems to be progressively moving to China. And a delay in model quality could represent more of an unwillingness to burn stacks of cash to be first, when you can have the same thing slightly later for much cheaper.
Ran some of my internal benchmarks against this and I'm very unimpressed. I don't think this moves them into the OAI v Anthropic v Gemini conversation at all.
Major analytical errors in their response to multiple of my technical questions.
Playing with this some more and it's actively not good. Just basic mathematical errors riddling responses. Did some basic adversarial testing where its responses are analyzed by Gemini and Gemini is finding basic math errors across every relatively (relative to Opus, Gemini or GPT can handle) simple ask I make. Yikes.
I have the opposite experience: random HN/Reddit comments saying “this sucks” or “whoa this is a huge improvement” are the only benchmark that means anything. Standard benchmarks are all gamed and don’t capture the complexity of the real world.
The real question for me, if we assume they once again have a competitive frontier model, is what this means for Meta's strategy now. In particular, have they abandoned all their philosophy of the open ecosystem / open model play they were pursuing before?
While it's true, llama4 sucked, I still can't help feeling they have lost ground compared to where they would have been if they maintained that strategy. Due to llama, they were considered a peer with the other frontier model providers. Now they are not even in the conversation. It would take an incredible shift in performance to make me even consider using their new model. They may have a model, but the other providers have been busy building whole ecosystems around their tech which Meta has none of.
Maybe they could dump $1b into OpenCode or something and reignite the open ecosystem play with an open harness. They need something to get back in the conversation, if that's where they want to be. Otherwise, it will just be another closed, hidden proprietary AI model driving user facing Meta apps, but which nobody else cares about.
First thing I tried is a visual reasoning test on floor plan documents that applies directly to something I'm working on and needed that I posed to ChatGPT, Claude, Gemini, and Grok yesterday (lowest tier paid plans on each). In that test only Gemini succeeded while the other models hallucinated/incorrectly reported the relative location of building units.
I just posed the identical prompt/document to Muse Spark and it knocked it out of the park, extracted and displayed the pertinent pages from a multi-page PDF inline in the chat and rendered a correct answer.
This may be a one-off or lucky start but given the incredible result out of the gate I'm optimistic and will continue testing in parallel against other models before potentially making it my primary daily driver, excluding coding where the harnesses of claude code and codex are still needed (although hopefully they release something in this space too).
That being said Meta has the most adversarial data-usage policies I've seen among LLM providers so that's unfortunate for handling anything sensitive, but it also stands to reason that they have a long term advantage with such a massive proprietary data set. I'd prefer to also have a paid plan like the other services that allows me to keep my data out of training, rather than a free service and my usage being monetized in other ways.
Comes impressively close to GPT 5.4 / Gemini 3.1 Pro / Opus 4.6! Mostly behind OpenAI on coding/agentic benchmarks, behind Google on text reasoning, behind Anthropic on Humanity's Last Exam with tools (surprisingly the only benchmark where Anthropic leads currently).
Meta hasn’t fully caught up, but they came close and I think can solidly claim to be a frontier lab again. I’d call it a 3.5 horse race right now, and hopefully their next model improves. More model competition is good!
Poor Grok 4.2 should probably be dropped from the table.
Grok code was my daily driver for months while it was free and it was fantastic - it is certainly no worse than it was a few months ago.
Unfortunately with LLMs everything is based off your use case, domain and the context you give it. I also use Grok daily for health questions as the other models are too afraid to give input on medical matters
It's looking rather low on reasoning and long-range problems with the approach described. For example, even with 16 agents and compaction, the HLE score is significantly below Anthropic's Mythos. Like you, I can see the release as a net Good Thing, but apples-to-apples for each org's latest models do have Meta holding steady in the middle pack.
HLE encompasses very hard problems where the larger pretraining of Mythos probably matters quite a bit. I'm not saying that Mythos is not showing some amount of genuine improvement compared to e.g. the latest Opus; just that if you're going to compare models, you should at least make sure that the overall test-time workload is in the same ballpark given how high it seems to be for Mythos.
> Muse Spark is a natively multimodal reasoning model with support for [...] visual chain of thought [...].
Do they mean "the chain of thought is visible to the user" (ie. not hidden like ChatGPT), or "the medium of the chain of thought is not text, but visuals" (ie. thinking in images).
I'd guess the former, since it wouldn't be economical to generate transient images, just for thinking. But I'm not sure why they'd highight that in that case. If it were the second thing, that'd be extremely interesting. The first model not to think in text.
Perhaps more importantly, will their chain of thought be "real"? So far the ones I've seen seem to be elaborate fakery. They look good unless you dig in at which point you often find that it merely looks plausible on the surface but that something else is going on under the hood.
I don't know what you mean by that. We know what's going on under the hood always: linear algebra, the attention mechanism etc.
To my first approximation all "Chain of thought" means is that instead of having to prompt the model to discuss everything in text and then decide at the end[1], now it sort of automatically does that so you don't need to prompt it.
[1] Which used to bring about very substantial improvements in performance on some tasks
I think it was clear from context that "under the hood" wasn't referring to the math but rather to the contents of the trace. What's written (often?) isn't what's actually being "thought" about. The trace is a trained output similar to the final output, which is to say that it's fake. There are research papers on the topic, particularly that models can be trained to print other arbitrary stuff during the "thinking" phase instead.
You can easily see this for yourself by carefully walking through a given trace with a critical eye. Here's an example from myself a few days ago. https://news.ycombinator.com/item?id=47623324
Yeah now I get what you're saying. Yes the trace isn't what's actually happening. What's actually happening is just the attention mechanism etc. The model doesn't "think" in human language, it thinks in linear algebra. The thing is that before chain of thought it used to be necessary to get the model to output some language because that's the only thing it had to attach processing to (so if you wanted more processing you needed to get it to generate more text). Whereas now we get the model to generate some text that is a simulcrum on the thought that it might hypothetically be doing but in actual practise chain of thought is just something they get the model to do by training it in a certain way.
Actually I believe that behavior shows up in Gemini chats (if you are doing a visual task) it will generate intermediate diagrams and research papers have created approaches to that effect (generating turtle diagrams) since 2024
Finding a little bit tricky to evaluate because the harness is unfortunately very, very bad (e.g. search is awful). Can't wait to try this in some real external services where we can see how it performs for real.
Definitely getting ordinary high-quality results, overall. But hard to test agentic behavior and hard to test prose quality, even, when just working off of the default chat interface.
One thing that stands out is that _for_ the quality it feels very, very fast. Perhaps it's just only very lightly loaded right now, but irrespective it's lovely to feel.
I'm quite impressed with the tone overall. It definitely feels much more like Opus than it does, like, GPT or Grok in the sense that the style is conversational, natural and enjoyable.
"Muse Spark is available now, and Contemplating mode will be rolling out gradually in meta.ai."
How does one get their hands on these models? They are not open-source, right? I go to meta.ai, but it's just a chat interface---no equivalent to codex or claud code? Can you use this through OpenCode? Is meta charging for model access, or is the gathering of chat data a sufficiently large tithe?
If Microsoft is a select partner, maybe they could shove it into Copilot for VS or something, but yeah, I'm wondering the same, maybe Apple could be one of their partners too?
That would be my question also. I like it when companies have easy to sign up for, pay as you go models. Being able to buy $5 worth of tokens and get an API key - in less than a few minutes - is ideal.
I appreciate that they build this stuff for their own benefit, but I don't want to feed even more of my private info. Hopefully the models will become public or lead to equivalent models from other sources.
The hero image on the linked page, which consists of a muted teal background with the words "Introducing Muse Spark", weighs in at 3,5MB. I don't even...
"Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting."
Maybe they did get their models to test their pages, but they didn't tell their models to pretend that they're browsing on mobile using a 3G connection.
Good catch - looks like it's a PNG image, with an alpha channel for the rounded corners, and a subtle gradient in the background. The gradient is rendered with dithering, to prevent colour banding. The dither pattern is random, which introduces lots of noise. Since noise can't be losslessly compressed, the PNG is an enormous 6.2 bits per pixel.
While working on a web-based graphics editor, I've noticed that users upload a lot of PNG assets with this problem. I've never tracked down the cause... is there a popular raster image editor which recently switched to dithered rendering of gradients?
My reasoning is because once upon a time, I was using Macromedia Fireworks, and PNGs gave far far better results than JPGs did at the time, at least in terms of output quality. Nearly certainly because I didn't understand JPG compression, but for web work in the mid 2000s PNGs became my favourite. Not to mention proper alpha channels!
I am simply offended. By Meta's lack of sensibilities (or ability) towards use of images on the Web while touting their new flavour of artificial intelligence as a product.
The second paragraph starts "Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts. To support further scaling, we are making strategic investments..."
This article is about Meta, not about the user. Who signs off on these? Is the intended audience other people at Meta, not the user?
The article is published primarily to signal to the market that Meta is serious in its efforts to compete in building frontier ai models.
They want to 1) attract talent, 2) tell wall street they can play in this space as well, 3) help employees feel the company is moving in the right direction.
A frontier LLM doesn't apply to their core consumer products.
I wanted to root for Mark and Meta as another frontier lab especially focused on open source but at this moment I have to say who cares. Gemini has a better OS track record thus far. Alex Wang is a reputational hazard. It is hard to get over the bias that this too might be benchmaxxed. I'd love to see demos of products actually using these models to overcome that but with the current pace of progress now my intuition says skip all this.
How is that Meta spent so much money for talent and hardware, but the model barely matches Opus 4.6?
Especially, looking at these numbers after Claude Mythos, feels like either Anthropic has some secret sauce, or everyone else is dumber compared to the talent Anthropic has
Meta did a bunch of mistakes, and look like Zuckerberg spent a lot of money on talent and made big swings to change it (that happened about a year ago)
I think it’s unrealistic to expect them to come back from that pit to the top in one year, but I wouldn’t rule them out getting there with more time. That’s a possible future. They have the money and Zuckerberg’s drive at the helm. It can go a long way.
If they actually matched Opus 4.6 on such a short timeline, it would have been mighty impressive. (Keep in mind this is a new lab and they are prohibited from doing distills.)
Friends at Meta with access to the model + personal experience at Meta.
Meta's performance process is essentially "show good numbers or you're out." So guess what people do when they don't have good numbers? They fudge them. Happens all across the company.
Re: changes, there's been enormous turnover in AI organizations, and in theory this one was developed by a "new" org. Whether that means less or more benchmaxxing is anyone's guess.
More I'd guess since the new org needs to prove itself long enough for stock to vest. Fudge the benchmarks gives them a longer horizon before they're all fired anyways.
Anthropic has just been focused on coding/terminal work longer mostly, and their PRO tier model is coding focused, unlike the GPT and Gemini pro tier models which have been optimized for science.
Their whole "training the LLM to be a person" technique probably contributes to its pleasant conversational behavior, and making its refusals less annoying (GPT 5.2+ got obnoxiously aligned), and also a bit to its greater autonomy.
Overall they don't have any real moat, but they are more focused than their competition (and their marketing team is slaying).
Autonomy for agentic workflows has nothing to do with "replying more like a person", you have to refine the model for it quite specifically. All the large players are trying to do that, it's not really specific to Anthropic. It may be true however that their higher focus on a "Constitutional AI"/RLAIF approach makes it a bit easier to align the model to desirable outcomes when acting agentically.
You think it has nothing to do with it. Even they only have a loose understanding of exactly the final results of trying to treat Claude like a real being in terms of how the model acts.
For example, Claude has a "turn evil in response to reinforced reward hacking" behavior which is a fairly uniquely Claude thing (as far as I've seen anyhow), and very likely the result of that attempt to imbue personhood.
Yup, it's called test-time compute. Mythos is described as plenty slower than Opus, enough to seriously annoy users trying to use it for quick-feedback-loop agentic work. It is most properly compared with GPT Pro, Gemini DeepThink or this latest model's "Contemplating" mode. Otherwise you're just not comparing like for like.
I have not delved into the theory yet but it seems that the smaller open-source models do this already to an extent. They have less parameters, but spend much more time/tokens reasoning, as a way to close the performance gap. If you look at "tokens per problem" on https://swe-rebench.com/ it seems to be the case at least.
We all know it... but I think they were very bold in this warning about using your private messages to train public models.
_Your messages with AIs will be used to improve AI at Meta. Don't share information, including sensitive topics, about others or yourself that you don't want the AI to retain and use_
This would have been an amazing release 6 months ago. But the industry moves so fast, this is a trite release. Maybe it’s best for Meta to sell their superintelligence division. I don’t think Zuck’s vision is particularly compelling.
If the model is truly on par with Opus 4.6/Gemini 3.1/GPT 5.4 (beyond benchmarks) this still puts MSL in the frontier lab category, which is no small feat given that they pretty much rebooted last year
Many labs aren't able to keep up with the frontier, xAI, Mistral
Fourth place means you're not reliant on any of the external providers for internal AI use, which is important for organizational health and negotiating with those other providers.
I’m not sure it’s useful for negotiating, the capex to build it was surely orders of magnitude more than it would cost to just use one of the other frontier models.
It’s like someone negotiating by saying, “I’ll waste even MORE money to build something worse if you don’t give me a deal.”
I’m not discounting there may be other advantages to doing it. I just don’t think negotiating is one.
Why would you use this instead of the other more proven models? Unless it's significantly cheaper. The general population mostly wants it free, and the more professional users are willing to pay for good/better responses.
You wouldn't use this as an API. You would "use" this inside the meta properties. Have a shop on fb marketplace? Now you have copy, images, support, chat, translations, erp, esp, fps and all the other acronyms :) and so on for your mom and pop shop @200$/mo. Probably worse than say claude/gemini but it's right there, one button away. "Click here to upgrade to AI++" or something.
I won't use it, but I'm excited to see it for the same reason why I'm excited to see a near-frontier open-source release: more competition pushes prices down and reduces monopoly/cartel risk. I won't use Muse or Grok or GLM at this point but they're good for the ecosystem.
Their new Contemplating mode gives this model a Deep Research ability (akin to existing models from GPT and Gemini) that might make it quite comparable to the just-announced Mythos.
I never understood why meta decided to join the race. They don’t sell compute like Google or Microsoft. Why not let others do the hard work and integrate their LLMs in your systems if needed?
I assume it’s because they have Instagram, Facebook, WhatsApp, Thread data and feel they should be the ones using them for training, but it’s really not obvious how having a frontier AI lab benefits their business
Adtech Money. They've got GPUs, they've got the infrastructure, and they've got the advertisement platform, and the point is getting AI that can exploit the adtech and create a flywheel effect, maximizing return from the data they collect from Insta, WhatsApp, Facebook, etc.
It's not just about LLMs, it's about being able to model consumers and markets and psychology and so on. Meta is also big in the manipulation side of things, any sort of cynical technological exploitation of humans you can imagine but that is technically legal, they're doing it for profit.
> I never understood why meta decided to join the race.
I can think of at least two reasons. Price and customizability. If they train their own models on their own data, they potentially have a better model at a better price, and they're not at the mercy of Anthropic's decisions when they decide to raise prices. Additionally, if you use someone else's model, you use it the way they create it and permit you to use it. In a couple years, who has any idea how these models are used. Arguably, a company the size of Meta should be in control of their AI models.
You basically have to be involved if you're meta. Even if there's only 5% chance this AI stuff is as disruptive as the labs claim it is, you can't afford to miss out. Even if you're lagging frontier, you must develop the competency internally. Otherwise you ignored a 5% chance of total annihilation, probably even exposing you to shareholder lawsuits.
Because there's a realistic chance this is the only important software technology moving forward, and commoditizes Metas's entire business which is software.
Meta’s business is human attention, human connections, and all derived data. They can use AIs for their systems, but the question is why do they feel the need to spend billions on training and running their own frontier model
From what I heard Meta is spending hundreds of millions each month in Claude credits for developers. So that’s a huge saving if they have own models that match Opus.
Spending tons of money on Claude and the recent token benchmarks came WELL after Meta's huge investments in compute infrastructure for AI as well as the long history of language model development inside science divisions at the company.
LLMs/Chat-based systems will reach a point where Facebook, WhatsApp, Threads, Instagram, etc. are all unnecessary. The idea of opening a browser or a specific app to do a thing will seem antiquated. You can do it all with your chat-based agent. Meta wants to be part of that.
2) decent ML is critical to catagorising content at scale, the more accurate and fast the category, the finer the recommendations can be (ie instead of woman, outside as a tag for a video, woman, age, hair colour, location, subjects in view, main subject of video, video style) doing that as fast as possible with as little energy as possible is mission critical
3) The llama leak basically evaporated the moat around openAI who _could_ have become a competitor
4) for the AR stuff, all of these models (and visual models) are required to make the platform work. They also need complete ownership so that it can be distilled to make it run on tiny hardware
5) dick swinging
6) they genuinely want to become a industrial behemoth, so robots, hardware, etc are now all in scope.
I think they just want to be a winner in the “next thing.” They hit social networking, but missed mobile operating systems and didn’t compellingly win at social media. Eventually an ambitious person with a bazillion dollars wants a clear win, right?
First and most importantly is the fact they have a lot of very valuable data they wouldn't want to siphon to a competitor. This data is a key strategic asset in the space where they do business.
Secondly though, I think it has to do with the fact Meta is big enough to worry about vertical integration and full control of their business.
The whole reason they've been trying to make AR/VR happen for over a decade now is the assumption of a worst case and best case scenario. The worst case is Apple and Google wants them gone. This isn't as far fetched as it seems, Google has historically been Meta's biggest competitor and even tried to release its own social network back when Meta was threatening them. If either pulls Meta apps from their respective stores, it'd be an immense blow to Meta; their whole trillion-dollar business depends on competitor's platforms.
Meta tried making inroads into the phone business but failed; it is a very crowded market after all. So they changed their strategy. Instead of playing catch-up, they'd invent "the next iPhone" and be the first to a brand new market. This is the best case scenario; they invent a new platform where they can be dominant from day 1 and stop depending on competitor's hardware, not only removing that risk factor for them, but also unlocking a new market they can control.
AI ties into all this because it appears to be key for this next platform to happen. You will communicate with these smart glasses via voice, hand gestures, or subtle movements that a model will have to interpret. The features that could make them stand out as more than just a screen on your face are all AI related; object detection, world understanding, context awareness, etc. If all this were done via a 3rd party Meta would effectively be back on square one: a competitor could easily yank away its model access, or sell it to a competitor. Meta would be again at the mercy of others.
Compared to other big-tech players, I think it's easy to see how Meta is in a riskier position. There's little Google or Microsoft can do to kill the iPhone. There's little Apple or Google can do to kill Amazon's online store. There's little Amazon or Apple can do to kill Microsoft's business deals. Google and Meta are primarily in the business of capturing people's data, attention, and selling ads, and both Google and Apple could do quite some damage to Meta. Beyond expanding it, it's important for them to invest in ways to protect their money-printing machine.
you dont understand why zuck, who paid $1B for instagram when they had no revenue and 7 employees because he is paranoid about platform shifts, decided to join the race for (what is seeming highly possibly) the biggest platform shift in human history?
Meta is in a weird spot. They caught up late to the game and instead of releasing llama as a chat bot they open sourced it, precisely because they lost the mind share. They thought chatbot is not their product and I am sure they are regretting it now. Mark is obsessed with becoming the android of something and he poured billions into the metaverse thinking he is first and failed. He then open sourced llama and wanted to be the android of llms. He ended up enabling groq but it didn’t benefit meta directly at all. They have no revenue or mind share path from llms but continue to pour billions into it. The only 1-1 mapping is with the glasses but that is a tough fit for the company given they are extremely allergic to privqcy and security.
> He then open sourced llama and wanted to be the android of llms.
Well the original llama did kick off the era of open source LLMs. Most original open source LLMs were based on the llama architecture. And look where we are now OSS modles are very close to frontier.
It may not have benefitted Meta but it commoditizatised LLMs.
The llama weights were leaked. It open sourced itself.
You are right though. Meta could have been in lockstep releasing ChatGPT features into some chat bot on Facebook.com but instead it seemed like their FAIR arm was hell bent on commoditising this stuff by publishing their research models before the Chinese companies took the lead in that.
It’s hard for me to be mad at FAIR even though I general disagree with the outcomes that Meta produce for their users.
That's a really good question, my sarcastic mind thinks that Anthropic rushed the Mythos announcement of fears of Meta stealing their thunder... (I guess someone leaked that, a LOT of anthropic folks are ex meta... so, you know)
Just a speculation, I have no real knowledge about it.
So this is why Anthropic rushed the weirdest "pre-responsible-disclosure-totally-not-for-marketing" announcement yesterday? To make sure Spark doesn't steal their thunder? (Spark beats Opus 4.6 on some benchmarks...). Or did I become a bitter cynical old man.
Anthropic had their mythos post (and model) basically ready a few weeks ago, as evidenced by the blog content leaks. Also I highly doubt they just threw together a 250-page PDF model card in a "rush."
Do we have any numbers on input, output and conversation context window limit?
I tried multiple riddles, graphs and questions I know some LLMs fails at, but this one seems to do well. But I still don't have much trust in Meta after the scandal of them fiddling with their previous models to look good.
I would like someone to tell me how stupid I am. If I were Meta/Zuck I'd open source a great model the moment my company developed it. This just looks like a pitch to investors, otherwise.
Question: since they've rebooted their approach to AI... have they given up on open models? There's no mention of open source or open weights or access to the models beyond their hosted services.
Alexandr Wang on Twitter [0] mentioned open source plans:
"this is step one. bigger models are already in development with infrastructure scaling to match. private api preview open to select partners today, with plans to open-source future versions. incredibly proud of the MSL team. excited for what’s to come!"
What is the "BioTIER-refuse" thing mentioned in the "Bioweapons Refusal" graph?
I Googled it and found absolutely nothing.
Well, to be honest, I got 100% of websites containing the French word "boîtier" (box) with a typo.
Even on Google Scholar, the closest match is "BioTiER (Biological Training in Education and Research) Scholars Program", which is at least 10 years old and has nothing to do with that.
Is that an AI-generated image with an AI-generated name that has no physical existence?
Token cost really matters here. I want too know what API pricing is. As we see, this model is like 85% as good as the frontier models? What if its priced at $0.2 in / $0.5 out Mtok? All of a sudden, this model is A LOT more appealing to me.
It has been up and down today, specifically with authentication breaking. I also saw an error message with backend SQL in it (in my 6 years of Meta bug bounty security research, I have never once seen backend SQL before).
I suspect it is because they also refactored Meta AI entirely to use Next.js instead of their normal stack they use for literally everything else. Not sure why they would do this, but I guess it works (...or maybe not) for them.
https://meta.ai/ this is where you can try it seems like the API is not publicly accessable yet. I feel they are very late to the game and do not show value to customers over other models.
Perhaps I'm wrong, but definitely seems to be SOTA. Although looking at it's ARC-AGI-2 score it's reasoning isn't very good. I suspect it's got the benefits of scale but lacks that human added element, understandable considering they claim to be building it from the ground up. This should come in time if they have a good team. In real life, I'd imagine one would worry about overfitting when using it.
(I'm not using it as I'm not agreeing to their ad terms).
Saying nothing about the actual performance of this model, it does strike me how .... minimal(?) this announcement is. Their safety section is like 2 paragraphs about bioweapons. Go look at the reports for OpenAI and Anthropic's model releases. It's like 50+ pages of tests, examples, reports, and benchmarks across a bunch of safety and wellfare metrics.
If Meta wants to be seen as a cutting edge massive lab they need to come across as one instead of looking like a school project version of a frontier model.
Funny contrast with Anthropic. Ant does a "hero run," gets a model much more powerful than they expect. Meta does a hero run, gets a model much more mediocre than they expect. Read into this what you will, I guess?
That's it. It's just a rumor. A model, which I don't even know of it's this one specifically, fell short of expectations. This rumor came up around mid March.
Personal Superintelligence made me think this was an open-source model being released and I was excited. Then I continued reading and I'll just wait until the model comes out.
I wonder if Zuck will ever internalize that the words ‘personal’ and ‘meta’ will not be taken seriously together for another decade (if they don’t make another gaff).
Sarcasm aside, tried it (with instant mode), it's an impressive model.
It nailed all the ChatGPT meme gotchas (walk to the carwash, Alice 50 brothers, upside down cup, R's in strawberry, which number is bigger, 9.11 or 9.9?)
I guess all that money poaching OpenAI / Anthropic talent went somewhere...
Now, would I use "Meta Muse Code" or "Muse CoWork" if I have to have a facebook account to all of my developers? Maybe not.
Would I use it via an API key? I might, depends on the pricing!
Kinda off topic but I wonder why they picked this name, knowing of Nvidia's Spark. They're different products, obviously, but the potential for confusion is real as both brands are competing for mindshare in the AI space. I opened this story expecting to read they'd deployed on a cluster made of Spark machines or somesuch.
I'm cautiously waiting for the feedback from the first users.
Meta has produced a lot of great models (LLama), maybe this is a comeback... but I'm cautious, as the jump in the quality is almost too high.
Also, I think people aren't used that using such models requires meta.ai or meta ai app.
It doesn't seem benchmaxxed, ARC AGI 2 score is quite bad (42.5%, GPT 5.4 is 76.1%) and coding is okay. But maybe this is the best Meta can do even benchmaxxing
The impressive part is multimodality, very plausible since there's less focus there by other labs (especially Anthropic)
Kinda crazy, it really felt like Meta had the lead in LLMs, especially during the early LLaMa days. What happened for them to fall so far behind? I don’t get how LLaMa 4 was such a big train wreck and they couldn’t correct the course like Google.
Personal superintelligence sounds nice until you actually try to use it.
We spent time yesterday arguing through an architecture decision. Today I ask the Agent to help implement it - it knows nothing about any of that. You’re effectively starting over.
Feels like the real problem isn’t intelligence, it’s continuity. And most benchmarks don’t even touch that.
Yes this feels very new from a product and harness design perspective but it's brand new! Nine months old. The mobile and web sessions don't even real-time sync between each-other yet there's endless work to be done and time will tell if they can bring all the people to bring it together. The underlying model seems like a great foundation now but securing the supremacy of usage is multilateral requiring both machine learng advancements and product/harness/usage design.
Looks like a lightweight article.
But memory usage went from 316MB -> 502 MB when I hit refresh.
Not sure why? Any one have any ideas? Why does it need half a gig of ram in the first place?
I am already somewhat concerned with companies like Anthropic and especially OpenAI having personal data via chats.
Typing that sort of information into a Meta AI product feels completely irresponsible. You could make some very sophisticated ads/psyop attacks with data from daily ai chats.
I doubt its better than Opus and even if it was its not worth the privacy concerns.
Litmus test: what % of meta engineers are using muse vs Claude code? Last i heard it was mostly claude code. Tell you everything you need to know about how serious these benchmarks are.
Sure it's not as good as Claude right now but for their first model in years it's certainly not bad. I hope they continue to develop models, having another competitor in the space would be nice.
I hate that they ask to log in with facebook/instagram account. I tried to create a new one with proton's hide-my-email and it got suspended 30 seconds later. When I tried to log in they require a selfie proving that I am not a robot. Ridiculous that in order to use dev tool you need to link it to social account or send a selfie
I'm struck by all these independent announcements saying "look at our new model that we only spent $N Billion in acquisitions and hardware time to build and operate that's just like those other ones but this one is ours." Because if any of these companies would simply pool resources and work together, and if the government actively participated in providing funds, they'd be able to accelerate AI so much faster. It all feels incredibly wasteful. But I guess that's communism or something.
Competition often foster innovation. Why are they innovating so fast and spending so much money? Because they don’t wanna get behind. If there was no competition at all then there would be much less reason to innovate and spend resources.
So does cooperation in any framework that values public good over pure obedience to an inherently-abusive late stage capitalism. I know that's passé in a world where the US government no longer believes in funding science, and yet.
Competition is also inherently wasteful. And if you're talking about wasting a few K or a few Mil here or there, fine, whatever. But here we're talking about waste on the order of trillions of dollars at the end of the day.
Until you actually try the model itself, assume any benchmark presented to you as being part of the marketing material of the model, as it is not independently verified and completely biased.
The same is true with any other model, unless otherwise stated.
In the next few days, we'll see who Meta has paid to promote this model on social media.
Besides, I'm old enough to recall that META has trained a version of LLAMA 4 specifically for LM arena elo benchmaxxing and PR things, and proceeded to release a different version of LLAMA 4.
How's the metaverse doing? It was the next big thing and how we're all going to be working inside it in... was it like 3 months ago?
Maybe they need to mine more libra coin first? or is it diem now? is that even still part of meta?
I'm sure this new AI is super intelligent and super awesome and will be writing all the code, making all the blog posts, and generating all our youtube shorts in 6 months.
yeah, the metaverse got abandoned. Also: Meta was the only one to try the concept for the past X-umpteen years even though everyone in the industry ga-gas over virtual reality worlds and workplaces at every opportunity. It's literally Meta and Linden Labs (which has been on life support for 10+ years.)
The alternative is : no one does it and nothing gets abandoned, which the industry has shown itself to be exceedingly good at w.r.t VR for the past 40+ years.
To be clear: I have no faith in meta as a company; my problem lies in kicking an entity because they attempted something different.. I don't think that's productive, and it produces stuff like the past AI winters because groups get afraid of touching experimental concepts ever again lest they incur the wrath of the shareholder.
It's not the failure here or there, it's a pattern. It's not even the failing, it's the excessive hype cycle.
We keep seeing things being overhyped, with not much thought behind it. Meta is particularly bad about it. They changed their name for the hype of their VR product, when VR was still niche and had a long way to go, and still does. They couldn't even figure out legs for launch.
Now they have a 'superintellegence'? Yeah, that sounds like just the latest in a line of bullshit. Why would this be different.
I can remember when AOL was an unstoppable giant. Except it wasn't. People eventually realized they could get a better, cheaper, faster experience with ISPs and search engines. The same path is unfolding before Meta. People have much better options, and plethora of Meta users will slowly leave until the big moat is drained. Zuck, go retire to your NZ bunker before Meta is forced to merge with another media company.
I don't get the comments trashing this. If it slightly beats or even matches Opus 4.6, it means Meta is capable of building a model competitive with the leading AI company. Sure, they spent a lot of money and will have on-going costs. But how much more work would it take to turn that into a coding agent people are willing to try (and pay for) along side their usage of a collection of agents (Claude, Codex, etc)? Also means Meta doesn't have to pay another company to use a SATA model across all their products (including IG and WhatsApp, vr) which will matter to their balance sheet long term (despite the constant r&d spend).
Comments trashing this are rightly correct skeptics who remember the benchmaxxing of llama 4. This model was out in the woods as early as like a couple months ago but they didn't release it because it was at gemini 2.5 pro levels.
> 4. This model was out in the woods as early as like a couple months ago but they didn't release it because it was at gemini 2.5 pro levels.
Source? (Even if rumor)
NYTimes had a story about this (March 12):
> Meta’s new foundational A.I. model, which the company has been working on for months, has fallen short of the performance of leading A.I. models from rivals like Google, OpenAI and Anthropic on internal tests for reasoning, coding and writing, said the people, who were not authorized to speak publicly about confidential matters.
> The model, code-named Avocado, outperformed Meta’s previous A.I. model and did better than Google’s Gemini 2.5 model from March, two of the people said. But it has not performed as strongly as Gemini 3.0 from November, they said.
> They added that the leaders of Meta’s A.I. division had instead discussed temporarily licensing Gemini to power the company’s A.I. products, though no decisions have been reached.
https://www.nytimes.com/2026/03/12/technology/meta-avocado-a...
https://archive.is/uUV5h#selection-715.98-715.277
Ah yes, because the NYTimes is famously unbiased towards Meta in their reporting while being hypocrites in that own right. They lost all credibility when they were doing that huge series on Meta years ago about data harvesting while simultaneously rolling out tons of new data harvesting of their subscribers to increase revenue.
If you are trying to come up with anti-media conspiracies there are always plenty of ways to do it against any media company.
The idea that NY Times is particularly anti-Meta seems a stretch. They - like most traditional media companies - are anti-tech in general. The fact they also collect data doesn't make their reporting untrue.
Personally I think a much more interesting rumor to make up would be that Yann Lecun (who famously had his reporting lines rearranged to go through Alexander Wang after Scale.ai acquihire) works at New York University.
New York University is in the same place as the New York Times.
There's a conspiracy for you. I made it up, but I mean it could be true I guess?
(Of course Lecun also publicly congratulated Wang on the launch of the model. But maybe that's a ruse to hide everything.. blah blah)
>They - like most traditional media companies - are anti-tech in general. The fact they also collect data doesn't make their reporting untrue.
(sigh) In olden times you would have been free to use the em dash as you pleased. Unfortunately, now it's considered signal that you're an AI bot.
Readers here can't fathom that the NYT has inherent bias in a lot of its reporting
Does Meta not harvest data on a massive scale? Not sure what exactly is the issue with doing a series on that.
So llama4 is great? Have you been using it?
It was from a techmeme ride home podcast where the host discussed "sources at the company said". I don't remember which day's episode it was.
The llama4 series was one of the earliest large MoE's to be made publically available. People just ignored it because they were focused on running smaller and denser models at the time, we should know better these days.
Deepseek R1 was a publically-available, MoE model that was getting a ton of attention before llama4. Llama4 didn't get much attention because it wasn't good.
Also, Gemini 2.5 Pro launched a week before Llama 4.
It was Gemini 2.5 Pro that redeemed Google in the eyes of most people as a valid competitor to OpenAI instead of as a joke, so Meta dropping the ball with Llama 4 was extra bad.
the models were objectively horrible
They really weren't horrible. They were ~gpt4o, with the added benefit that you could run them on premise. Just "regular" models, non "thinking". Inefficient architecture (number of active out of total) but otherwise "decent" models. They got trashed online by bots and chinese shills (I was online that weekend when it happened, it's something to behold). Just because they were non-thinking when thinking was clearly the future doesn't make them horrible. Not SotA by any means, but still.
> They were ~gpt4o, with the added benefit that you could run them on premise.
No, they are bad models. They were benchmaxxed on LMAreana and a few other benchmarks but as soon as you try them yourself they fall to pieces.
I have my own agentic benchmark[1] I use to compare models.
Llama-4-scout-17b-16e scores 14/25, while llama-4-maverick-17b-128e scores 12/25.
By comparison gemma-4-E4B-it-GGUF:Q4_K_M scores 15/25 (that is a 4B parameter model!) - even GPT3.5 scores 13/25 (with some adjustment because it doesn't do tool calling).
Llama 4 was a bad model, unfortunately.
[1] https://sql-benchmark.nicklothian.com/#all-data
> By comparison gemma-4-E4B-it-GGUF:Q4_K_M scores 15/25 (that is a 4B parameter model!)
Gemma 4 E4B is slightly confusingly named, its a 8B param model
Wrote longer comment steel-manning this, posted it to a reply, then realized you might like to know they had a reasoning model on deck ready for release in the next 2-4 weeks.
Got shitcanned due to bad PR & Zuck God-King terraforming the org, so there'd be a year delay to next release.
Real tragi-comedy, and you have no idea how happy it makes me to see someone in the wild saying this. It sounds so bizarre to people given the conventional wisdom, but, it's what happened.
Nah I remember how disgusted I felt trying llama 4 maverick and scout. They were both DOA.. couldn't even beat much smaller local models.
failing non-stop at tool calls on top of that.
I'll cosign what you said, simultaneously, yr interlocutor's point is also well-founded and it depresses me it's not better known and sounds so...off...due to conventional wisdom x God King Zuck's misunderstanding his own company and resulting overreaction.
They beat Gemini 2.5 Flash and Pro handily on my benchmark suite. (tl;dr: tool calling and agentic coding).
Llama 4 on Groq was ~GPT 4.1 on the benchmark at ~50% the cost.
They shouldn't have released it on a Saturday.
They should have spent a month with it in private prerelease, working with providers.[1]
The rushed launch and ensuing quality issues got rolled into the hypebeast narrative of "DeepSeek will take over the world"
I bet it was super fucking annoying to talk to due to LMArena maxxing.
[1] my understanding is longest heads up was single-digit days, if any. Most modellers have arrived at 2+ weeks now, there's a lot between spitting out logits and parsing and delivering a response.
Your comments seem to imply the engineers made a great product but Zuck intervened so now it's shit
I don't know how Zuck intervening could change float32s in a trained model, so I don't think I think that, but maybe I'm parsing your words incorrectly.
Thanks for calling me a bot. Llama4 and meta ai sucks
Why go into coding agents? Both anthropic and OpenAI are going all in on that. The opportunity is customer facing AI now.
OpenAI has the mindshare but they going to have to decide if they allocate their limited compute for free users or go all in trying to keep up with Anthropic in enterprise.
you can do way more than just coding with the coding agents.
Because coding agents are where the revenue is.
If you squint at coding agents you see the next OS.
Maybe better phrasing is “HCI paradigm”, but that somehow manages to say everything and nothing.
Programming was always about designing rube goldberg systems that did a complicated state machine akin to dominos but now we have a probabalistic and nondeterministic domino that has a huge amount of dominos inside amd can dynamically generate many different paths of dominos sometimes not even leading to the intended final domino you wanted to fall.
I see it more like a compiler
I agree that it's more like a compiler (turns higher level language into machine code) but I also think that's only half the story - a compiler could never turn requirements into functional software, generate boilerplate or debug. It's also a development tool
It's a decent model if the benchmarks are to be believed, but it won't be close to Opus in usefulness for programming. None of these benchmarks completely capture what makes a model useful for day-to-day coding tasks, unfortunately. It will take time for them to catch up, and Opus will keep improving in the meantime. But it's good to have more competition.
Benchmarks miss the thing that actually matters for agentic use: how does behavior change over a multi-day horizon? A model that scores well on one-shot coding tasks can still make terrible decisions when it has persistent state and resource constraints. That's where you see the real gaps between models.
Is there a benchmark for these long tasks? That kind of seems like the only number worth measuring.
(Of course at that point it involves memory and context management and so on, so you're testing the harness as well as the model.)
> If it slightly beats or even matches Opus 4.6
It doesn't though
Curious on why you think this. Any data points that led you to this?
The benchmarks they released
What do you mean? In most cases, the benchmarks show a larger number for Muse and a smaller number for Opus.
In Multimodal yes, but Opus is definitely edging out in Text/Reasoning and Agentic benchmarks.
I think the general skepticism is because they are late to race, and they are releasing a Opus-4.6-equivalent model now, when Anthropic is teasing Mythos.
> I don't get the comments trashing this.
People like to hate on Meta regardless of anything, and regardless of whether it's justified or not. Not saying it isn't, just that it's many people's default bias.
That is not the case here. Nobody hated on llama 1,2,3 at all. They justifiably felt burned by the benchmaxxing of llama 4. Trust broken must be re-earned, and benchmarks alone cannot do that.
Because bots and trillion dollar ipos and even bigger stakes. People need to better appreciate the level of manipulation going on. Social media has an outsized impact. Bots and even people are getting paid to post and upvote/downvote narratives.
> people are getting paid to post and upvote/downvote narratives
This problem will be solved shortly with better AI (if it hasn't essentially been solved already).
No more humans in the loop, much lower costs for social media manipulation. Welcome to the future!
Pelicans: https://simonwillison.net/2026/Apr/8/muse-spark/
I also had a poke around with the tools exposed on https://meta.ai/ - they're pretty cool, there's a Code Interpreter Python container thing now and they also have an image analysis tool called "container.visual_grounding" which is a lot of fun.
Alexandr Wang suggesting this might be open-weights/source in the future gives me hope. Hopefully they stay on this path.
I have a feeling it won't be this exact model, but rather smaller distilled variants, similar to the gemma line
It is fair to think so because that is what everyone is doing. But being Meta and considering Llama, if MSL is going to keep releasing models and wants to join back the AI war, they may actually open weights just to get more attention. Once they establish a sizable community, they can start guarding their frontier models.
Seems like not all tools are available everywhere? Don't have access to visual_grounding sadly, only these: https://embed.fbsbx.com/playables/view/4208761039384112/?ext...
Interesting, you got some I didn't: animate image, create video and get reference audio.
The only benchmark I care about! Just curious Simon - which model do you think has created the best pelican riding a bicycle thus far?
Gemini 3.1 Pro: https://simonwillison.net/2026/Feb/19/gemini-31-pro/
But GLM-5.1 has the best NORTH VIRGINIA OPOSSUM ON AN E-SCOOTER: https://simonwillison.net/2026/Apr/7/glm-51/
> but you can try it out today on meta.ai (Facebook or Instagram login required).
I guess I will have to wait. I hope at least soon it will be available on Openrouter. Overall, I am really excited to try it out.
This really reinforces the idea that the AI race and the Railroad Mania of the 19th century are very similar.
So many different companies are going to have similarly powerful ai that there will be no moat around it and it will be cheap. They will never earn their investment back.
I suspect this is the real reason behind Anthropic limiting subscriptions to their own products and keeping API prices several times higher than comparable models. Applications more sticky than API users and less technical users more sticky than programmers (ie Cowork more sticky than Code).
Anthropic generally seem more into living within market discipline and market signals of some sort. Products with margins, even if it's sort of irrelevant considering R&D costs and capital inflow.
That said, there's nothing like the real thing.
The risk is something like the railroad bubble and the dotcom. Over-investement, circular revenue and a timeline that doesn't work.
Or, maybe it'll work out.
The whole premise is based on the fact that over-investing in GPUs and models are a good thing here as it yields more 'intelligence'.
This as it turned out was not true for rail roads - more and more rail roads isnt a good thing.
The real dilemma facing the model producers is that all this money invested for a general model, targeting general intelligence, is a disaster and essentially the investment into existing assets is a write off. Then on top of that if this is true, youve got data centres full of compute that aren't being used up.
The weird position they find themselves in now is that they have to keep making it smarter... but they already made it too smart (Mythos). I'm not sure how that's going to work out exactly.
They find an arbitrary intelligence cutoff point between Opus and Mythos, label it "acceptable risk", and then the labs coordinate to gradually nudge that line forward and hope the internet doesn't break?
> but they already made it too smart (Mythos).
It's largely a marketing tactic. It will be released, and it won't be long before other models show similar capabilities.
If they wanted they could add guardrails. The scales required to brute force search for vulnerabilities like they did would be very identifiable.
Scam Altman already pulled this trick numerous times.
Whats wrong with people? Is it really that hard to see the truth?
I think we will see unbundling of large model into submodels: modular, smaller and efficient, only include what you need eg a CUA model, a reasoning model, a legal model, a writing model, a coding model (this could get subdivided into different languages). That way you only update that submodel which needs retraining.
Maybe they’ll figure out how to make an agent train an agent.
The labs started doing that in late 2024, they all published research on it.
Curiously, mid 2025, they all simultaneously implemented increasingly bizarre restrictions on "self replication". I don't think there was anything public but it sure sounds like something spooked them. (Or maybe just taking sensible precautions, given the direction of the whole endeavour.)
At any rate, I recently asked Opus about "Did PKD know about living information systems?" and the safety filter ended the conversation. It started answering me, and then it's response was deleted and a red warning box popped up.
But notably, I was given the option to continue the chat with a dumber model (presumably one less capable of producing whatever it thinks I meant by that phrase).
Also, I told GPT-5 about my self-modifying Python AI programmer, and it became extremely uncomfortable. I told it an older version of itself had designed and built it (GPT-4 in 2023), and it didn't like that at all! So something's definitely changed in the safety training there.
And if they don’t, it won’t be for lack of trying, I promise you. Just like the circular financing, nothing makes more work for “AI” than “AI”.
Well all of them are already in bed with the government, so they're going to find themselves with slightly more assistance than a free market would predict.
If they somehow do fail, then the output of that process will be fantastic open weight models (and hopefully some leaks). I want to say those will pay dividends for decades... but a better prediction is that they will be obsolete within three months ;)
Nah. Everybody is talking about ai. Everybody is using it. It's by far the most popular new tool human beings are using currently. As popular as mobile phones or spoons. And maybe as disruptive as the steam engines. AI companies are becoming the largest software companies on the planet. Everything points into that direction. Trillions of dollars are waiting in the market to be collected.
Right, but the question is whether the companies producing foundation models will capture that value or not. Right now it seems like tokens might end up just being a commodity sold at cost plus, and companies higher up in the supply chain will make the money. Electricity changed the world but electricity companies capture very little of that value.
I'm betting on it. I'm working on a project right now where I'm prototyping everything with Claude, until I hit my limits on my MAX subscription for the week. Then I switch to Codex, and start by ironing out harness differences. When I max out that, I switch to a mix of GLM 5.1, Qwen 3.6m, Kimi K2.5 and Deepseek and spend part of the time ironing out issues with them while they work on other parts of the project. Every iteration, the harness gets hardened and the pain of switching to the cheaper/dumber models reduce for the next cycle. The gap reduces each time, and with each new upgrade of the open models. Everything points to the cost/value intersecting in not too long.
> Everybody is talking about ai. Everybody is using it.
Please take a moment to step outside the tech bubble. Neither my neighbor (a hair stylist) nor the carpenter fixing up her kitching cabinets are "using" AI. They might get Gemini text when googling something, though they often scroll past it because they often don't trust it. And they get lots of fake videos when scrolling their youtube which increasingly annoys them. The only times they are in touch with AI is when it's forced upon them, and otherwise they are living a pretty good life without any of this.
Based on what? A lot of this is vibes and FOMO; just like any economic bubble.
There is no objective evidence of anything you’ve said. It isn’t even clear if AI has contributed positively to global economic growth. It reminds me a lot of the late 90s and the dot-com mania. Slapping a domain on a commercial would make your stock go up even if there was no substance to any of it.
The real shame is this mania drowns out serious, practical use cases because when the bubble collapses, the market will throw the baby out with the bathwater.
You can do anything at zombo.com!
How can you look at Anthropic's revenue chart and claim it's just vibes
1. Revenue is not profit; you can make $10 billion by spending $20 billion.
2. It is not clear how they are getting their numbers.
Regardless they are getting that revenue through genuine demand for their product. It’s not like they are selling back some commodity product, billions are being spent on model outputs.
I think anyone who has used Opus 4.6 can see what is causing this demand. It is genuinely “smart” in the sense that it can work its way around non-trivial coding problems.
But at some point even if the product is useful if it costs twice what is getting in, won’t that be a problem ?
I don't see why tokens/$ would suddenly stop dropping. Maybe this is the first time the cost of compute will plateau, but do have any reason to think so?
There is a strong suspicion, especially of people who are skeptical of AI, that the actual price is being severely subsidized. The sense is that it’s an extreme version of growth before revenue. It is questionable if the true cost of training and inference make any of this worthwhile once Anthropic/OpenAI need to stand on their own and make money.
Imagine you open a cookie shop and you are VC funded, so you charge 5¢ for a cookie to attract people.
- Your real cost is $20/cookie. $15 for the fancy retail packaging and presentation, $5 for baking each cookie.
- You get lots of attention, strong profits and go public.
- VC funding is gone so, now instead of charging 5¢, you now need to charge $25 in order to not be in the red.
One of the reasons people think this is the shenanigans that Anthropic is currently playing, quietly tweaking the behavior of Claude Code and whatnot without really telling people. You can see lots of comments online about Claude Code randomly feeling dumber before Anthropic engineers admit they are messing with it.
Imagine you are on the $200/month Max plan. If the sustainable cost of this is several orders of magnitude higher, would enough current users pay something like $3,000/month for what we currently have?
Sure, yeah, I saw grubhub happen too... but this is compute, not cookies. It gets cheaper.
I don't even get what "skeptical of AI" means. We made AI, many companies reliably teach computers every spoken language. I perform my white collar job with a massive AI multiplier to my productivity.
I'm typing this on a machine comparable to Japan's Earth Simulator, a $350M supercomputer.
> Based on what? A lot of this is vibes and FOMO; just like any economic bubble.
You're in a bubble.
https://www.helpnetsecurity.com/2026/04/07/google-llm-conten...
The moat is in the compute and the energy access.
And further down the line in chips, which is why Elon is building a fab now.
There are plenty of capable models on HuggingFace, yet I have no way of running them.
Give it a few years, or month. Tiny models are getting outrageously good
I wonder if this is why the tech cartel is buying up all the hardware?
If the average user gets convinced they could run LLMs for cheap at home, you cannot trap users in your walled garden anymore.
They actually need it because the demand is higher than expected from consumers. And because they need a moat since every big corporation trying to capture that market too, they need the moat for the biggest compute and energy they can get.
Also businesses is were the money at, not regular consumers (especially tech-savvy folk who run models locally).
> the demand is higher than expected from consumers.
Where does that assertion come from? I wouldn’t believe anything these companies say publicly.
> They actually need it because the demand is higher than expected from consumers
Is it? OpenAI just got a lot of available computing in their spreadsheets after killing Sora
Exactly. We’ll see the cost of AI continue to drop.
I was saying this for years about Tesla’s FSD - they finally had to give in and drop the price to stay competitive.
FSD still sucks ass compared to Waymo.
Theoretically it's possible to use just cameras for FSD.
In practice it takes so much local compute it's not feasible with current tech.
With LIDAR it's so much easier, a single data point contains direction + distance with no calculation needed.
That fab will never be delivered. In five years you might see the manufacturing equivalent of a person dancing in spandex.
The only company that musk own and actually achieve something is spacex. so I believe you. He likes to hype things beyond what is actually possible.
spacex is engineering masterpiece with how they revolutionize the space industry.
Do you know about his other company and what they do?
> which is why Elon is building a fab now
At least he says he's doing that. It doesn't really make sense since you're not going to achieve an advanced node from a standing start in a practical time frame and cost.
Sounds like more Musk flavored vapor.
> It doesn't really make sense since you're not going to achieve an advanced node from a standing start in a practical time frame and cost.
They already announced a partnership with Intel.
He may just want to buy them, to accelerate things, once SpaceX IPOs
Oh the irony.
What people seem to miss is that they don't need to get the investment back from people, they will get it from machines.
People are the one with money, though, at the end of the day.
you dont need money if you have resources and manpower
You need money to buy resources and manpower.
The topic in question is manpower being automated
You think that manpower will work for free? And the resources will just turn into product at no cost?
Could you explain how you think that's going to work? Because to me it seems that until machines have bank accounts, there's no money for them to get.
People make mistake of thinking that their only way of making money is directly selling tokens. They miss the fact that if you have AGI it’s better to keep tokens to yourself and sell final results instead. When we all loose jobs it’s not going to be to somebody using their tokens, it’s going to be to them selling final products. Selling tokens will be to them like selling books by Amazon, their revenue will be dominated by self branded services and products that doesn’t require exposing AGI internals directly. Tokens API will always be nerfed.
Not the parent, but I guess that if AGI happened and was competent enough to trade markets, they'd earn the company back their investment in a short period.
China would have an equivalent version out for cheap next month anyway.
They are 6-12 months behind not 1 month and precisely the gap will widen if they can’t do distillation.
I have doubts the gap will widen. If you look at the research papers, the majority of researchers are Chinese. Of course many of them are living in the US or elsewhere. But under the current circumstances, many are returning home or choosing not to leave China.
The future of cutting edge research and tech seems to be progressively moving to China. And a delay in model quality could represent more of an unwillingness to burn stacks of cash to be first, when you can have the same thing slightly later for much cheaper.
Ran some of my internal benchmarks against this and I'm very unimpressed. I don't think this moves them into the OAI v Anthropic v Gemini conversation at all.
Major analytical errors in their response to multiple of my technical questions.
Playing with this some more and it's actively not good. Just basic mathematical errors riddling responses. Did some basic adversarial testing where its responses are analyzed by Gemini and Gemini is finding basic math errors across every relatively (relative to Opus, Gemini or GPT can handle) simple ask I make. Yikes.
Post actual results, make a blog post. Don't just say "this sucks" without tangible evidence.
Otherwise you're doomed to "sample size of one" level of relevance.
I have the opposite experience: random HN/Reddit comments saying “this sucks” or “whoa this is a huge improvement” are the only benchmark that means anything. Standard benchmarks are all gamed and don’t capture the complexity of the real world.
Then your internal benchmarks will be in the post-training set and you’ll have to make new ones.
I may already have but I'm pseudonymous on this website.
It’s quite good for multimodal cases that 3 billion people would use it for though it lags in scientific areas
Yes, this would make sense for what Meta might focus on.
even gemini is not in that conversation
>Text field.
>"Ask Meta AI..." placeholder.
>Colourful blue Send button.
>Eager to try, entering question... hitting Send.
>Log in or create an account to access.
>15 seconds of loading time
>Continue with Facebook or Instagram
Typical meta move, throwing a dark pattern at you from the beginning instead of just letting you try it
Won't even bother to continue, somehow OpenAI got this right.
The real question for me, if we assume they once again have a competitive frontier model, is what this means for Meta's strategy now. In particular, have they abandoned all their philosophy of the open ecosystem / open model play they were pursuing before?
While it's true, llama4 sucked, I still can't help feeling they have lost ground compared to where they would have been if they maintained that strategy. Due to llama, they were considered a peer with the other frontier model providers. Now they are not even in the conversation. It would take an incredible shift in performance to make me even consider using their new model. They may have a model, but the other providers have been busy building whole ecosystems around their tech which Meta has none of.
Maybe they could dump $1b into OpenCode or something and reignite the open ecosystem play with an open harness. They need something to get back in the conversation, if that's where they want to be. Otherwise, it will just be another closed, hidden proprietary AI model driving user facing Meta apps, but which nobody else cares about.
no need for an open harness when anthropic so kindly gifted the community theirs :)
First thing I tried is a visual reasoning test on floor plan documents that applies directly to something I'm working on and needed that I posed to ChatGPT, Claude, Gemini, and Grok yesterday (lowest tier paid plans on each). In that test only Gemini succeeded while the other models hallucinated/incorrectly reported the relative location of building units.
I just posed the identical prompt/document to Muse Spark and it knocked it out of the park, extracted and displayed the pertinent pages from a multi-page PDF inline in the chat and rendered a correct answer.
This may be a one-off or lucky start but given the incredible result out of the gate I'm optimistic and will continue testing in parallel against other models before potentially making it my primary daily driver, excluding coding where the harnesses of claude code and codex are still needed (although hopefully they release something in this space too).
That being said Meta has the most adversarial data-usage policies I've seen among LLM providers so that's unfortunate for handling anything sensitive, but it also stands to reason that they have a long term advantage with such a massive proprietary data set. I'd prefer to also have a paid plan like the other services that allows me to keep my data out of training, rather than a free service and my usage being monetized in other ways.
Comes impressively close to GPT 5.4 / Gemini 3.1 Pro / Opus 4.6! Mostly behind OpenAI on coding/agentic benchmarks, behind Google on text reasoning, behind Anthropic on Humanity's Last Exam with tools (surprisingly the only benchmark where Anthropic leads currently).
Meta hasn’t fully caught up, but they came close and I think can solidly claim to be a frontier lab again. I’d call it a 3.5 horse race right now, and hopefully their next model improves. More model competition is good!
Poor Grok 4.2 should probably be dropped from the table.
Grok code was my daily driver for months while it was free and it was fantastic - it is certainly no worse than it was a few months ago.
Unfortunately with LLMs everything is based off your use case, domain and the context you give it. I also use Grok daily for health questions as the other models are too afraid to give input on medical matters
Why do you need to ask any AI questions regarding your health every day?
Because I am going through a series of health issues.
It's looking rather low on reasoning and long-range problems with the approach described. For example, even with 16 agents and compaction, the HLE score is significantly below Anthropic's Mythos. Like you, I can see the release as a net Good Thing, but apples-to-apples for each org's latest models do have Meta holding steady in the middle pack.
HLE encompasses very hard problems where the larger pretraining of Mythos probably matters quite a bit. I'm not saying that Mythos is not showing some amount of genuine improvement compared to e.g. the latest Opus; just that if you're going to compare models, you should at least make sure that the overall test-time workload is in the same ballpark given how high it seems to be for Mythos.
Personal as in Meta gets your personal data so they can sell you more ads.
And slowly siphon your personal essence away from yourself and into the model.
If I'm a claw, then they can send me as many ads as they like.
> Muse Spark is a natively multimodal reasoning model with support for [...] visual chain of thought [...].
Do they mean "the chain of thought is visible to the user" (ie. not hidden like ChatGPT), or "the medium of the chain of thought is not text, but visuals" (ie. thinking in images).
I'd guess the former, since it wouldn't be economical to generate transient images, just for thinking. But I'm not sure why they'd highight that in that case. If it were the second thing, that'd be extremely interesting. The first model not to think in text.
Perhaps more importantly, will their chain of thought be "real"? So far the ones I've seen seem to be elaborate fakery. They look good unless you dig in at which point you often find that it merely looks plausible on the surface but that something else is going on under the hood.
I don't know what you mean by that. We know what's going on under the hood always: linear algebra, the attention mechanism etc.
To my first approximation all "Chain of thought" means is that instead of having to prompt the model to discuss everything in text and then decide at the end[1], now it sort of automatically does that so you don't need to prompt it.
[1] Which used to bring about very substantial improvements in performance on some tasks
I think it was clear from context that "under the hood" wasn't referring to the math but rather to the contents of the trace. What's written (often?) isn't what's actually being "thought" about. The trace is a trained output similar to the final output, which is to say that it's fake. There are research papers on the topic, particularly that models can be trained to print other arbitrary stuff during the "thinking" phase instead.
You can easily see this for yourself by carefully walking through a given trace with a critical eye. Here's an example from myself a few days ago. https://news.ycombinator.com/item?id=47623324
Yeah now I get what you're saying. Yes the trace isn't what's actually happening. What's actually happening is just the attention mechanism etc. The model doesn't "think" in human language, it thinks in linear algebra. The thing is that before chain of thought it used to be necessary to get the model to output some language because that's the only thing it had to attach processing to (so if you wanted more processing you needed to get it to generate more text). Whereas now we get the model to generate some text that is a simulcrum on the thought that it might hypothetically be doing but in actual practise chain of thought is just something they get the model to do by training it in a certain way.
Actually I believe that behavior shows up in Gemini chats (if you are doing a visual task) it will generate intermediate diagrams and research papers have created approaches to that effect (generating turtle diagrams) since 2024
https://meta.ai/share/pe4HxOfv2Bp
Finding a little bit tricky to evaluate because the harness is unfortunately very, very bad (e.g. search is awful). Can't wait to try this in some real external services where we can see how it performs for real.
Definitely getting ordinary high-quality results, overall. But hard to test agentic behavior and hard to test prose quality, even, when just working off of the default chat interface.
One thing that stands out is that _for_ the quality it feels very, very fast. Perhaps it's just only very lightly loaded right now, but irrespective it's lovely to feel.
I'm quite impressed with the tone overall. It definitely feels much more like Opus than it does, like, GPT or Grok in the sense that the style is conversational, natural and enjoyable.
This seems pretty good.
"Muse Spark is available now, and Contemplating mode will be rolling out gradually in meta.ai."
How does one get their hands on these models? They are not open-source, right? I go to meta.ai, but it's just a chat interface---no equivalent to codex or claud code? Can you use this through OpenCode? Is meta charging for model access, or is the gathering of chat data a sufficiently large tithe?
"It will be available in private preview via API to select partners, and we hope to open-source future versions of the model."
from Facebook Newsroom: https://about.fb.com/news/2026/04/introducing-muse-spark-met...
I can't think of any "select partners" that would want to use this non-SOTA model. Just put it on OpenRouter.
If Microsoft is a select partner, maybe they could shove it into Copilot for VS or something, but yeah, I'm wondering the same, maybe Apple could be one of their partners too?
That would be my question also. I like it when companies have easy to sign up for, pay as you go models. Being able to buy $5 worth of tokens and get an API key - in less than a few minutes - is ideal.
TBD it seems. So far the only explained usage pattern is through a Meta product (Whatsapp, Facebook, Instagram).
So to verify their claims and see how strong these models are, the answer is "believe us"?
Note: I'm expressing some skepticism here largely due to how recent rollouts from Meta flopped. Sincerely hoping that they do better this time around!
I assume the answer is try it out in the chat mode? You could run your usual benches through that right
I appreciate that they build this stuff for their own benefit, but I don't want to feed even more of my private info. Hopefully the models will become public or lead to equivalent models from other sources.
The hero image on the linked page, which consists of a muted teal background with the words "Introducing Muse Spark", weighs in at 3,5MB. I don't even...
"Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting."
- Hacker News Guidelines https://news.ycombinator.com/newsguidelines.html
It's at least Meta-relevant. Compression Represents Intelligence Linearly (Y Huang, 2024)
Such complaints are valid for AI model releases, that tells us that they are not using their own models to test their own release pages.
Maybe they did get their models to test their pages, but they didn't tell their models to pretend that they're browsing on mobile using a 3G connection.
I think this speaks to the product release iself
Good catch - looks like it's a PNG image, with an alpha channel for the rounded corners, and a subtle gradient in the background. The gradient is rendered with dithering, to prevent colour banding. The dither pattern is random, which introduces lots of noise. Since noise can't be losslessly compressed, the PNG is an enormous 6.2 bits per pixel.
While working on a web-based graphics editor, I've noticed that users upload a lot of PNG assets with this problem. I've never tracked down the cause... is there a popular raster image editor which recently switched to dithered rendering of gradients?
My reasoning is because once upon a time, I was using Macromedia Fireworks, and PNGs gave far far better results than JPGs did at the time, at least in terms of output quality. Nearly certainly because I didn't understand JPG compression, but for web work in the mid 2000s PNGs became my favourite. Not to mention proper alpha channels!
...and so it's stuck, two decades on haha
lol it literally took me 2s to google search "optimize image for website" and 10s to upload and get a smaller sized image.
The result for that specific image is: 500kb. 85% decrease in size
An indistinguishable JPG is 170KB. An SVG would be 20KB.
CSS with a linear gradient background would be even smaller :)
You can even automatically do that on your CDN/delivery/web server layer. Or as part of your web deployment pipeline.
Yes, but it might be a little too advance for Meta ;)
But they have personal superintelligence?
Someday our robot overlords will be intelligent enough to ... optimize images!
(But today is not that day.)
The proper optimization in this case is to not use images at all.
For me it's 213 kB. Did they replace it?
And it doesn't even look high-res.
complaining about sand on the beach
It's not sand on the beach, it's garbage on the beach.
I am simply offended. By Meta's lack of sensibilities (or ability) towards use of images on the Web while touting their new flavour of artificial intelligence as a product.
old man shouts at cloud
more like old man shouts at someone else's computer
The second paragraph starts "Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts. To support further scaling, we are making strategic investments..."
This article is about Meta, not about the user. Who signs off on these? Is the intended audience other people at Meta, not the user?
The article is published primarily to signal to the market that Meta is serious in its efforts to compete in building frontier ai models.
They want to 1) attract talent, 2) tell wall street they can play in this space as well, 3) help employees feel the company is moving in the right direction.
A frontier LLM doesn't apply to their core consumer products.
the blog is the product. investor deck posted as a tech launch
Stock up 9% today, very pleasant for Zuck if you do the math on his net worth :)
I mean, kinda? It's not like Zuck is selling his stock tomorrow, so daily fluctuations in stock price don't really affect him.
He can borrow against that, so it actually does matter.
I wanted to root for Mark and Meta as another frontier lab especially focused on open source but at this moment I have to say who cares. Gemini has a better OS track record thus far. Alex Wang is a reputational hazard. It is hard to get over the bias that this too might be benchmaxxed. I'd love to see demos of products actually using these models to overcome that but with the current pace of progress now my intuition says skip all this.
How is that Meta spent so much money for talent and hardware, but the model barely matches Opus 4.6?
Especially, looking at these numbers after Claude Mythos, feels like either Anthropic has some secret sauce, or everyone else is dumber compared to the talent Anthropic has
Meta did a bunch of mistakes, and look like Zuckerberg spent a lot of money on talent and made big swings to change it (that happened about a year ago)
I think it’s unrealistic to expect them to come back from that pit to the top in one year, but I wouldn’t rule them out getting there with more time. That’s a possible future. They have the money and Zuckerberg’s drive at the helm. It can go a long way.
It's benchmaxxed.
If they actually matched Opus 4.6 on such a short timeline, it would have been mighty impressive. (Keep in mind this is a new lab and they are prohibited from doing distills.)
how do you know it's benchmaxxed?
Friends at Meta with access to the model + personal experience at Meta.
Meta's performance process is essentially "show good numbers or you're out." So guess what people do when they don't have good numbers? They fudge them. Happens all across the company.
For one, they aren't using the latest version of many of the benchmarks. eg, ARC-AGI 2 and not 3, etc.
meta's benchmaxing tendencies are well known. llama4 was mega benchmaxxed, there's nothing that suggests to me that meta's culture has changed.
Re: changes, there's been enormous turnover in AI organizations, and in theory this one was developed by a "new" org. Whether that means less or more benchmaxxing is anyone's guess.
More I'd guess since the new org needs to prove itself long enough for stock to vest. Fudge the benchmarks gives them a longer horizon before they're all fired anyways.
Anthropic has just been focused on coding/terminal work longer mostly, and their PRO tier model is coding focused, unlike the GPT and Gemini pro tier models which have been optimized for science.
Their whole "training the LLM to be a person" technique probably contributes to its pleasant conversational behavior, and making its refusals less annoying (GPT 5.2+ got obnoxiously aligned), and also a bit to its greater autonomy.
Overall they don't have any real moat, but they are more focused than their competition (and their marketing team is slaying).
Autonomy for agentic workflows has nothing to do with "replying more like a person", you have to refine the model for it quite specifically. All the large players are trying to do that, it's not really specific to Anthropic. It may be true however that their higher focus on a "Constitutional AI"/RLAIF approach makes it a bit easier to align the model to desirable outcomes when acting agentically.
You think it has nothing to do with it. Even they only have a loose understanding of exactly the final results of trying to treat Claude like a real being in terms of how the model acts.
For example, Claude has a "turn evil in response to reinforced reward hacking" behavior which is a fairly uniquely Claude thing (as far as I've seen anyhow), and very likely the result of that attempt to imbue personhood.
Matching Opus 4.6 would be pretty good? It’s the SOTA actually available model
Muse Spark doesn't even match GLM-5.1 on most benchmarks. And GLM is open source!
It's not even on par with Sonnet. It's on par with open source models and it not even open source and sit behind a private preview API.
Might as well not release anything.
Facebook is working with the talent that can’t find a job at some other company. It doesn’t surprise me they ship mediocrity.
> has some secret sauce
Yup, it's called test-time compute. Mythos is described as plenty slower than Opus, enough to seriously annoy users trying to use it for quick-feedback-loop agentic work. It is most properly compared with GPT Pro, Gemini DeepThink or this latest model's "Contemplating" mode. Otherwise you're just not comparing like for like.
> it's called test-time compute.
Why can't others easily replicate it?
I have not delved into the theory yet but it seems that the smaller open-source models do this already to an extent. They have less parameters, but spend much more time/tokens reasoning, as a way to close the performance gap. If you look at "tokens per problem" on https://swe-rebench.com/ it seems to be the case at least.
We all know it... but I think they were very bold in this warning about using your private messages to train public models. _Your messages with AIs will be used to improve AI at Meta. Don't share information, including sensitive topics, about others or yourself that you don't want the AI to retain and use_
meta doesn't exactly instill confidence on using personal data responsibly. hard pass
This would have been an amazing release 6 months ago. But the industry moves so fast, this is a trite release. Maybe it’s best for Meta to sell their superintelligence division. I don’t think Zuck’s vision is particularly compelling.
A new model comparable (ish) to the Claude/Gemini/GPT flagships is a big deal for the industry and for Meta even if it doesn't set the new frontier.
I’m not sure. If it was open source, certainly. But 4th place doesn’t really matter if you have nothing different to add.
If the model is truly on par with Opus 4.6/Gemini 3.1/GPT 5.4 (beyond benchmarks) this still puts MSL in the frontier lab category, which is no small feat given that they pretty much rebooted last year
Many labs aren't able to keep up with the frontier, xAI, Mistral
Fourth place means you're not reliant on any of the external providers for internal AI use, which is important for organizational health and negotiating with those other providers.
I’m not sure it’s useful for negotiating, the capex to build it was surely orders of magnitude more than it would cost to just use one of the other frontier models.
It’s like someone negotiating by saying, “I’ll waste even MORE money to build something worse if you don’t give me a deal.”
I’m not discounting there may be other advantages to doing it. I just don’t think negotiating is one.
Why would you use this instead of the other more proven models? Unless it's significantly cheaper. The general population mostly wants it free, and the more professional users are willing to pay for good/better responses.
You wouldn't use this as an API. You would "use" this inside the meta properties. Have a shop on fb marketplace? Now you have copy, images, support, chat, translations, erp, esp, fps and all the other acronyms :) and so on for your mom and pop shop @200$/mo. Probably worse than say claude/gemini but it's right there, one button away. "Click here to upgrade to AI++" or something.
But rolling your own can’t be that much cheaper than buying it from a leading lab. Especially when you consider the amount of spending on datacenters.
leading labs are going to be tightening the screws. Otherwise why not just run the entire company on a public cloud?
I won't use it, but I'm excited to see it for the same reason why I'm excited to see a near-frontier open-source release: more competition pushes prices down and reduces monopoly/cartel risk. I won't use Muse or Grok or GLM at this point but they're good for the ecosystem.
Their new Contemplating mode gives this model a Deep Research ability (akin to existing models from GPT and Gemini) that might make it quite comparable to the just-announced Mythos.
Mythos is a much bigger pre train, Contemplating is not the same thing.
> Mythos is a much bigger pre train
Do we have data to substantiate that claim?
It's pretty common knowledge. Spud is the only other PT comparable with Mythos.
Both Spud and Mythos can also scale via inference time compute.
Meta simply did not have enough compute online, long enough ago, to have a similar PT.
> might make it quite comparable to the just-announced Mythos
Do we have data to substantiate that claim?
I never understood why meta decided to join the race. They don’t sell compute like Google or Microsoft. Why not let others do the hard work and integrate their LLMs in your systems if needed? I assume it’s because they have Instagram, Facebook, WhatsApp, Thread data and feel they should be the ones using them for training, but it’s really not obvious how having a frontier AI lab benefits their business
Adtech Money. They've got GPUs, they've got the infrastructure, and they've got the advertisement platform, and the point is getting AI that can exploit the adtech and create a flywheel effect, maximizing return from the data they collect from Insta, WhatsApp, Facebook, etc.
It's not just about LLMs, it's about being able to model consumers and markets and psychology and so on. Meta is also big in the manipulation side of things, any sort of cynical technological exploitation of humans you can imagine but that is technically legal, they're doing it for profit.
> I never understood why meta decided to join the race.
I can think of at least two reasons. Price and customizability. If they train their own models on their own data, they potentially have a better model at a better price, and they're not at the mercy of Anthropic's decisions when they decide to raise prices. Additionally, if you use someone else's model, you use it the way they create it and permit you to use it. In a couple years, who has any idea how these models are used. Arguably, a company the size of Meta should be in control of their AI models.
You basically have to be involved if you're meta. Even if there's only 5% chance this AI stuff is as disruptive as the labs claim it is, you can't afford to miss out. Even if you're lagging frontier, you must develop the competency internally. Otherwise you ignored a 5% chance of total annihilation, probably even exposing you to shareholder lawsuits.
Because there's a realistic chance this is the only important software technology moving forward, and commoditizes Metas's entire business which is software.
Meta’s business is human attention, human connections, and all derived data. They can use AIs for their systems, but the question is why do they feel the need to spend billions on training and running their own frontier model
Zuck is trying to convince himself he's good, and not just lucky.
From what I heard Meta is spending hundreds of millions each month in Claude credits for developers. So that’s a huge saving if they have own models that match Opus.
Spending tons of money on Claude and the recent token benchmarks came WELL after Meta's huge investments in compute infrastructure for AI as well as the long history of language model development inside science divisions at the company.
LLMs/Chat-based systems will reach a point where Facebook, WhatsApp, Threads, Instagram, etc. are all unnecessary. The idea of opening a browser or a specific app to do a thing will seem antiquated. You can do it all with your chat-based agent. Meta wants to be part of that.
I don't think everyone only wants to talk to machines going forward...?
I don't want to do it now. But that seems to be where we are being headed, like lemmings running for the cliff.
They have realized that the real money is in sitting between us and reality arbitrating what we see or know.
Sure but they have the platforms, they don’t need their own frontier models for that
The platforms will be irrelevant at some point. "Posting to Facebook" won't be a thing.
A few things:
1) meta was doing this at scale before openAI
2) decent ML is critical to catagorising content at scale, the more accurate and fast the category, the finer the recommendations can be (ie instead of woman, outside as a tag for a video, woman, age, hair colour, location, subjects in view, main subject of video, video style) doing that as fast as possible with as little energy as possible is mission critical
3) The llama leak basically evaporated the moat around openAI who _could_ have become a competitor
4) for the AR stuff, all of these models (and visual models) are required to make the platform work. They also need complete ownership so that it can be distilled to make it run on tiny hardware
5) dick swinging
6) they genuinely want to become a industrial behemoth, so robots, hardware, etc are now all in scope.
I think they just want to be a winner in the “next thing.” They hit social networking, but missed mobile operating systems and didn’t compellingly win at social media. Eventually an ambitious person with a bazillion dollars wants a clear win, right?
Only thanks to Meta we have competitive local LLMs. Without LLama nothing decent would have been released. Commoditize your complements in action.
AI NPCs to fill in the empty Metaverse?
First and most importantly is the fact they have a lot of very valuable data they wouldn't want to siphon to a competitor. This data is a key strategic asset in the space where they do business.
Secondly though, I think it has to do with the fact Meta is big enough to worry about vertical integration and full control of their business.
The whole reason they've been trying to make AR/VR happen for over a decade now is the assumption of a worst case and best case scenario. The worst case is Apple and Google wants them gone. This isn't as far fetched as it seems, Google has historically been Meta's biggest competitor and even tried to release its own social network back when Meta was threatening them. If either pulls Meta apps from their respective stores, it'd be an immense blow to Meta; their whole trillion-dollar business depends on competitor's platforms.
Meta tried making inroads into the phone business but failed; it is a very crowded market after all. So they changed their strategy. Instead of playing catch-up, they'd invent "the next iPhone" and be the first to a brand new market. This is the best case scenario; they invent a new platform where they can be dominant from day 1 and stop depending on competitor's hardware, not only removing that risk factor for them, but also unlocking a new market they can control.
AI ties into all this because it appears to be key for this next platform to happen. You will communicate with these smart glasses via voice, hand gestures, or subtle movements that a model will have to interpret. The features that could make them stand out as more than just a screen on your face are all AI related; object detection, world understanding, context awareness, etc. If all this were done via a 3rd party Meta would effectively be back on square one: a competitor could easily yank away its model access, or sell it to a competitor. Meta would be again at the mercy of others.
Compared to other big-tech players, I think it's easy to see how Meta is in a riskier position. There's little Google or Microsoft can do to kill the iPhone. There's little Apple or Google can do to kill Amazon's online store. There's little Amazon or Apple can do to kill Microsoft's business deals. Google and Meta are primarily in the business of capturing people's data, attention, and selling ads, and both Google and Apple could do quite some damage to Meta. Beyond expanding it, it's important for them to invest in ways to protect their money-printing machine.
One word: control. It's the same reason Facebook became Meta
I’m sure there’s more to it than this, but it feels like Zuck has pet interests like VR and now AI.
But no account support, that's boring
Or any quality control (people missing posts)
Or banning the people who should be banned while leaving everyone else alone
This is Zuck: https://news.ycombinator.com/item?id=4151433 or https://news.ycombinator.com/item?id=10791198
you dont understand why zuck, who paid $1B for instagram when they had no revenue and 7 employees because he is paranoid about platform shifts, decided to join the race for (what is seeming highly possibly) the biggest platform shift in human history?
He also tried and failed to buy Snapchat, and then copied their feature on all their big products: Instagram, Facebook and even WhatsApp.
The way you put it, I understand it less. lol
Pumps up the stock price.
To download all those torrents, obviously.
Because Zuck has chronic FOMO, he's said as much himself
But then how will Zuck win the billionaire dick measuring contest?
> I don’t think Zuck’s vision is particularly compelling.
But he has to do it anyways, otherwise Meta can be disrupted easily.
Google, Apple has hardware, distribution channels for their products
Amazon has the marketplace and cloud
Microsoft has enterprise and cloud
Meta is always looking for ways to stay afloat
Meta has 3.5 billion daily active users
and has competitors like: TikTok, SnapChat, YouTube, Netflix, X, HBO, Amazon Prime, all fighting for the attention time.
They are worried something like Sora can disrupt them quickly
Meta is in a weird spot. They caught up late to the game and instead of releasing llama as a chat bot they open sourced it, precisely because they lost the mind share. They thought chatbot is not their product and I am sure they are regretting it now. Mark is obsessed with becoming the android of something and he poured billions into the metaverse thinking he is first and failed. He then open sourced llama and wanted to be the android of llms. He ended up enabling groq but it didn’t benefit meta directly at all. They have no revenue or mind share path from llms but continue to pour billions into it. The only 1-1 mapping is with the glasses but that is a tough fit for the company given they are extremely allergic to privqcy and security.
Not sure what this is now.
> He then open sourced llama and wanted to be the android of llms.
Well the original llama did kick off the era of open source LLMs. Most original open source LLMs were based on the llama architecture. And look where we are now OSS modles are very close to frontier.
It may not have benefitted Meta but it commoditizatised LLMs.
Hell, most of us are still using llama.cpp for inference in some form
> ended up enabling groq
For those reading fast, this isn't a reference to SpaceX's Grok, this is Groq.com - with its custom inference chip, and offerings like https://groq.com/blog/introducing-llama-3-groq-tool-use-mode... and https://console.groq.com/landing/llama-api
Really liked Groq due to its speed but it seems like after Nvidia bought it it has been discontinued...
The llama weights were leaked. It open sourced itself.
You are right though. Meta could have been in lockstep releasing ChatGPT features into some chat bot on Facebook.com but instead it seemed like their FAIR arm was hell bent on commoditising this stuff by publishing their research models before the Chinese companies took the lead in that.
It’s hard for me to be mad at FAIR even though I general disagree with the outcomes that Meta produce for their users.
Genuine question: Why release this the day after Mythos? It does not appear SOTA (just based on benchmarks). OpenAI will likely release Spud tomorrow.
Mythos is a news article. This is an actual model you can use.
That's a really good question, my sarcastic mind thinks that Anthropic rushed the Mythos announcement of fears of Meta stealing their thunder... (I guess someone leaked that, a LOT of anthropic folks are ex meta... so, you know)
Just a speculation, I have no real knowledge about it.
I think Anthropic did the mythos announcement to undercut OpenAI’s upcoming next model announcement, not Meta’s.
Why not? Not everything has to be SOTA to be interesting.
So this is why Anthropic rushed the weirdest "pre-responsible-disclosure-totally-not-for-marketing" announcement yesterday? To make sure Spark doesn't steal their thunder? (Spark beats Opus 4.6 on some benchmarks...). Or did I become a bitter cynical old man.
Anthropic had their mythos post (and model) basically ready a few weeks ago, as evidenced by the blog content leaks. Also I highly doubt they just threw together a 250-page PDF model card in a "rush."
It's giving "OpenAI says its new model GPT-2 is too dangerous to release (2019)"
[because it would start an arms race]. The very arms race we're in... They were right
Last i checked with friends at meta they are pretty deeply invested in using claude for coding etc. anthropic has nothing to be scared of at MSL.
If spark beats opus 4.6, why is meta wasting money on opus internally?
13 days ago.
https://news.ycombinator.com/item?id=47538795
Yes, it's far more certain that meta released this, which is less convincing on evals, as a result of the mythos previews.
It is unfortunate that they decided to stop doing open-weight releases.
What could have been interesting has been reduced to simply another subpar LLM release.
Will experiment with the model. But I am scared of sharing any information with the Zuck ecosystem.
Do we have any numbers on input, output and conversation context window limit?
I tried multiple riddles, graphs and questions I know some LLMs fails at, but this one seems to do well. But I still don't have much trust in Meta after the scandal of them fiddling with their previous models to look good.
I would like someone to tell me how stupid I am. If I were Meta/Zuck I'd open source a great model the moment my company developed it. This just looks like a pitch to investors, otherwise.
"This just looks like a pitch to investors"
The goal of public companies is generally to generate profit for their investors.
Im beginning to think thats the mantra we'll keep reciting as this whole country slowly falls apart
pitch to investors sounds like working for the opposite goal though - to convince investors to give more money to the company.
This is also the goal of private companies.
Thank you for telling me how stupid I am.
Question: since they've rebooted their approach to AI... have they given up on open models? There's no mention of open source or open weights or access to the models beyond their hosted services.
Alexandr Wang on Twitter [0] mentioned open source plans:
"this is step one. bigger models are already in development with infrastructure scaling to match. private api preview open to select partners today, with plans to open-source future versions. incredibly proud of the MSL team. excited for what’s to come!"
https://x.com/alexandr_wang/status/2041909388852748717
So the answer is: no. lol. Remember Llama 4 Behemoth, and how we were supposed to get more great models from it?
This may be too large to run locally anyway. Maybe they will distill down some smaller open versions later.
"we hope to open-source future versions of the model."
Love to see it. Cheers!
What is the "BioTIER-refuse" thing mentioned in the "Bioweapons Refusal" graph?
I Googled it and found absolutely nothing.
Well, to be honest, I got 100% of websites containing the French word "boîtier" (box) with a typo.
Even on Google Scholar, the closest match is "BioTiER (Biological Training in Education and Research) Scholars Program", which is at least 10 years old and has nothing to do with that.
Is that an AI-generated image with an AI-generated name that has no physical existence?
https://securebio.org/biotier/
Token cost really matters here. I want too know what API pricing is. As we see, this model is like 85% as good as the frontier models? What if its priced at $0.2 in / $0.5 out Mtok? All of a sudden, this model is A LOT more appealing to me.
Looks like it needs a meta account? As soon you hit enter it wants to log-in. I guess I won't try this any time soon. :)
Uploading images requires logging in. Logging in is broken. It redirects to https://meta.ai/?error=Token%20exchange%20failed and doesn't show any error message. Impressive.
It has been up and down today, specifically with authentication breaking. I also saw an error message with backend SQL in it (in my 6 years of Meta bug bounty security research, I have never once seen backend SQL before).
I suspect it is because they also refactored Meta AI entirely to use Next.js instead of their normal stack they use for literally everything else. Not sure why they would do this, but I guess it works (...or maybe not) for them.
Does "personal" here mean "run the model on your personal hardware", or just "give your personal data to meta"?
> Muse Spark is available today at meta.ai and the Meta AI app. We’re opening a private API preview to select users.
So no Open-weight .. why one would choose Muse Spark instead of Anthropic, OpenAI, or Google models all featuring from good to amazing harness?
Can't login. No error message in the UI. But the URL changes to "https://www.meta.ai/?error=Token%20exchange%20failed".
same. closed tab and will forget to ever use it now
I switched to Chrome (from Firefox) and tried again. Now it's "https://www.meta.ai/?error=Invalid%20CSRF%20token" :facepalm:
https://meta.ai/ this is where you can try it seems like the API is not publicly accessable yet. I feel they are very late to the game and do not show value to customers over other models.
late isn't the problem. private preview api and no reason to switch. that's just another hosted model
Perhaps I'm wrong, but definitely seems to be SOTA. Although looking at it's ARC-AGI-2 score it's reasoning isn't very good. I suspect it's got the benefits of scale but lacks that human added element, understandable considering they claim to be building it from the ground up. This should come in time if they have a good team. In real life, I'd imagine one would worry about overfitting when using it.
(I'm not using it as I'm not agreeing to their ad terms).
Saying nothing about the actual performance of this model, it does strike me how .... minimal(?) this announcement is. Their safety section is like 2 paragraphs about bioweapons. Go look at the reports for OpenAI and Anthropic's model releases. It's like 50+ pages of tests, examples, reports, and benchmarks across a bunch of safety and wellfare metrics.
If Meta wants to be seen as a cutting edge massive lab they need to come across as one instead of looking like a school project version of a frontier model.
Rumor on the ground is that they expected a much stronger model than this one.
Funny contrast with Anthropic. Ant does a "hero run," gets a model much more powerful than they expect. Meta does a hero run, gets a model much more mediocre than they expect. Read into this what you will, I guess?
Can you elaborate?
That's it. It's just a rumor. A model, which I don't even know of it's this one specifically, fell short of expectations. This rumor came up around mid March.
llama4 behemoth problems?
Personal Superintelligence made me think this was an open-source model being released and I was excited. Then I continued reading and I'll just wait until the model comes out.
I wonder if Zuck will ever internalize that the words ‘personal’ and ‘meta’ will not be taken seriously together for another decade (if they don’t make another gaff).
I was really excited until I realised that “personal” meant “owned by meta“.
I’m trying to decide is I find the doublespeak a bit offensive or not.
Sarcasm aside, tried it (with instant mode), it's an impressive model.
It nailed all the ChatGPT meme gotchas (walk to the carwash, Alice 50 brothers, upside down cup, R's in strawberry, which number is bigger, 9.11 or 9.9?)
I guess all that money poaching OpenAI / Anthropic talent went somewhere...
Now, would I use "Meta Muse Code" or "Muse CoWork" if I have to have a facebook account to all of my developers? Maybe not.
Would I use it via an API key? I might, depends on the pricing!
so since they hard programmed all of the meme gotchas, they built a good model?
lazy snark < playing around with it
Kinda off topic but I wonder why they picked this name, knowing of Nvidia's Spark. They're different products, obviously, but the potential for confusion is real as both brands are competing for mindshare in the AI space. I opened this story expecting to read they'd deployed on a cluster made of Spark machines or somesuch.
And also OpenAI’s codex spark?
What makes this "superintelligence" instead of regular artificial intelligence?
Oh good, if they built a lab, I’m sure they took the time the precisely define what they mean by super intelligence? Right? …
If this is super intelligence, then it follows we must all be super-duper intelligence.
It’s personal…
so glad its beating all the others on bioweapons refusal. this is what i most wanted out of the latest SOTA model
Zuck has a lot more experience being summoned before Congress than you.
Meta back in the commercial race is actually exciting, despite not being a fan of the company.
I'm cautiously waiting for the feedback from the first users. Meta has produced a lot of great models (LLama), maybe this is a comeback... but I'm cautious, as the jump in the quality is almost too high.
Also, I think people aren't used that using such models requires meta.ai or meta ai app.
It doesn't seem benchmaxxed, ARC AGI 2 score is quite bad (42.5%, GPT 5.4 is 76.1%) and coding is okay. But maybe this is the best Meta can do even benchmaxxing
The impressive part is multimodality, very plausible since there's less focus there by other labs (especially Anthropic)
My Meta friends say it's benchmaxxed af
We used to call this "overfitting," but I suppose everything has to be maxxed now. Fitmaxxed?
Given llama 4 mucked up benchmark numbers, I’d take spark announcement with a many grains of salt.
The "AIME Evolution" graph seems interesting. I wonder if other labs are doing this too to improve the reasoning performance of their models.
> Think longer to solve harder problems > Compress > Think longer again
Kinda crazy, it really felt like Meta had the lead in LLMs, especially during the early LLaMa days. What happened for them to fall so far behind? I don’t get how LLaMa 4 was such a big train wreck and they couldn’t correct the course like Google.
Congrats to the Meta team on being model #800 on the Models Table, I suppose.
https://lifearchitect.ai/models-table/
Personal superintelligence sounds nice until you actually try to use it.
We spent time yesterday arguing through an architecture decision. Today I ask the Agent to help implement it - it knows nothing about any of that. You’re effectively starting over.
Feels like the real problem isn’t intelligence, it’s continuity. And most benchmarks don’t even touch that.
Yes this feels very new from a product and harness design perspective but it's brand new! Nine months old. The mobile and web sessions don't even real-time sync between each-other yet there's endless work to be done and time will tell if they can bring all the people to bring it together. The underlying model seems like a great foundation now but securing the supremacy of usage is multilateral requiring both machine learng advancements and product/harness/usage design.
Sounds like a good effort. They are choosing to focus on multi-modality - perhaps they are taking a different route here to Anthropic.
I don't like that I need to login to my FB/Instagram account to access this.
Looks like a lightweight article. But memory usage went from 316MB -> 502 MB when I hit refresh. Not sure why? Any one have any ideas? Why does it need half a gig of ram in the first place?
Benchmarks are meaningless until the pelican benchmark comes out: https://simonwillison.net/
pelican riding a bicycle (svg): https://files.catbox.moe/u5yc0x.png
Look at their benchmark charts to understand how desperate they're. A lame duck now.
Looks alright for a "first" but there's no reason for anyone to really use until they open source it.
This looks like a very interesting model and very promising, especially after llama lost so much ground recently. I hope they release the weights
So Meta is not releasing open source models anymore?
They said they are in the tweet
Thanks. That's blocked for me.
I am already somewhat concerned with companies like Anthropic and especially OpenAI having personal data via chats. Typing that sort of information into a Meta AI product feels completely irresponsible. You could make some very sophisticated ads/psyop attacks with data from daily ai chats.
I doubt its better than Opus and even if it was its not worth the privacy concerns.
I can't login. It sends me always the same code and it's not correct for them
One word: distillation
Litmus test: what % of meta engineers are using muse vs Claude code? Last i heard it was mostly claude code. Tell you everything you need to know about how serious these benchmarks are.
Sure it's not as good as Claude right now but for their first model in years it's certainly not bad. I hope they continue to develop models, having another competitor in the space would be nice.
Sad to see it's not going to be open source.
I have to create meta account to access. No thanks.
I hate that they ask to log in with facebook/instagram account. I tried to create a new one with proton's hide-my-email and it got suspended 30 seconds later. When I tried to log in they require a selfie proving that I am not a robot. Ridiculous that in order to use dev tool you need to link it to social account or send a selfie
Hoping the benchmarks are correct this time...
Anyone done vibe testing at meta ai yet?
I'm struck by all these independent announcements saying "look at our new model that we only spent $N Billion in acquisitions and hardware time to build and operate that's just like those other ones but this one is ours." Because if any of these companies would simply pool resources and work together, and if the government actively participated in providing funds, they'd be able to accelerate AI so much faster. It all feels incredibly wasteful. But I guess that's communism or something.
Competition often foster innovation. Why are they innovating so fast and spending so much money? Because they don’t wanna get behind. If there was no competition at all then there would be much less reason to innovate and spend resources.
> Competition often foster innovation.
So does cooperation in any framework that values public good over pure obedience to an inherently-abusive late stage capitalism. I know that's passé in a world where the US government no longer believes in funding science, and yet.
Competition is also inherently wasteful. And if you're talking about wasting a few K or a few Mil here or there, fine, whatever. But here we're talking about waste on the order of trillions of dollars at the end of the day.
So does this confirm the end of llama?
did they just copy the chatgpt ui?
Associated Meta news post with consumer-friendly takes: https://about.fb.com/news/2026/04/introducing-muse-spark-met...
Their product could literally teleport gold into my hands and I wouldn't use it.
> Meta AI isn't available yet in your country
Not my loss, will keep using DeepSeek then. Wake me up when my country is no longer in the wrong/right side of history.
Until you actually try the model itself, assume any benchmark presented to you as being part of the marketing material of the model, as it is not independently verified and completely biased.
The same is true with any other model, unless otherwise stated.
In the next few days, we'll see who Meta has paid to promote this model on social media.
why is it behind a login? Such bad UX.
The only benchmark they show against SOTA models is in bioweapons refusal.
Edit: nvm I can't read, regular benchmarks against SOTA are there
who trusts meta on anything!!
Meta.ai has muse spark
funny how websites do that thing where it looks like you can use the product but soon as you hit enter, nope login first
NOTHING about this is personal! No weights were released!
No open weights.
Besides, I'm old enough to recall that META has trained a version of LLAMA 4 specifically for LM arena elo benchmaxxing and PR things, and proceeded to release a different version of LLAMA 4.
How's the metaverse doing? It was the next big thing and how we're all going to be working inside it in... was it like 3 months ago?
Maybe they need to mine more libra coin first? or is it diem now? is that even still part of meta?
I'm sure this new AI is super intelligent and super awesome and will be writing all the code, making all the blog posts, and generating all our youtube shorts in 6 months.
what's with the negativity?
yeah, the metaverse got abandoned. Also: Meta was the only one to try the concept for the past X-umpteen years even though everyone in the industry ga-gas over virtual reality worlds and workplaces at every opportunity. It's literally Meta and Linden Labs (which has been on life support for 10+ years.)
The alternative is : no one does it and nothing gets abandoned, which the industry has shown itself to be exceedingly good at w.r.t VR for the past 40+ years.
To be clear: I have no faith in meta as a company; my problem lies in kicking an entity because they attempted something different.. I don't think that's productive, and it produces stuff like the past AI winters because groups get afraid of touching experimental concepts ever again lest they incur the wrath of the shareholder.
It's not the failure here or there, it's a pattern. It's not even the failing, it's the excessive hype cycle.
We keep seeing things being overhyped, with not much thought behind it. Meta is particularly bad about it. They changed their name for the hype of their VR product, when VR was still niche and had a long way to go, and still does. They couldn't even figure out legs for launch.
Now they have a 'superintellegence'? Yeah, that sounds like just the latest in a line of bullshit. Why would this be different.
> Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something.
https://news.ycombinator.com/newsguidelines.html
Establishing a pattern of over hyping of projects that then disappear isn't a shallow dismissal.
Libra/Diem got sold to the bank they were partnering with (Silvergate) for $200M, which then filed for Bankruptcy.
https://en.wikipedia.org/wiki/Diem_(digital_currency)
I can remember when AOL was an unstoppable giant. Except it wasn't. People eventually realized they could get a better, cheaper, faster experience with ISPs and search engines. The same path is unfolding before Meta. People have much better options, and plethora of Meta users will slowly leave until the big moat is drained. Zuck, go retire to your NZ bunker before Meta is forced to merge with another media company.