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.
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.
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.
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 overreaction and misunderstanding of his own company.
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 prerelease working with providers.[1]
The face-planting at launch was being contrasted against DeepSeek will take over the world overhype.
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 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.
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.
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.
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.
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.
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.
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.
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.
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?
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.
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_
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.
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?
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.
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.
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!"
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.
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.
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.
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)
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.
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.
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.
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 prerelease working with providers.[1]
The face-planting at launch was being contrasted against DeepSeek will take over the world overhype.
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.
The goal of public companies is generally to generate profit for their investors.
It doesn't though
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.
Many labs aren't able to keep up with the frontier, xAI, Mistral
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.
Do we have data to substantiate that claim?
Do we have data to substantiate that claim?
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.
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 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.
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.
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.
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
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
They are worried something like Sora can disrupt them quickly
Major analytical errors in their response to multiple of my technical questions.
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.
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?
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
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.
Might as well not release anything.
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.)
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.
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.
Why can't others easily replicate it?
"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
Also, I think people aren't used that using such models requires meta.ai or meta ai app.
The impressive part is multimodality, very plausible since there's less focus there by other labs (especially Anthropic)
Not my loss, will keep using DeepSeek then. Wake me up when my country is no longer in the wrong/right side of history.
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.
Edit: nvm I can't read, regular benchmarks against SOTA are there
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.
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.
https://news.ycombinator.com/newsguidelines.html
https://en.wikipedia.org/wiki/Diem_(digital_currency)