Meta to open source versions of its next AI models
Posted by abkibaarnsit@reddit | LocalLLaMA | View on Reddit | 58 comments
Posted by abkibaarnsit@reddit | LocalLLaMA | View on Reddit | 58 comments
Love_Cat2023@reddit
Before qwen 4?
brubits@reddit
Suuuuuuuuuuure
mindwip@reddit
Call me when it's released
throwawayacc201711@reddit
Sounds like they’re releasing the models and then down the road they’ll open source them. Sounds like a bunch of bs
Spitfire1900@reddit
TBF at the going rate of innovation SOTA models have no value after 8-12 months and should just be open sourced
DeepOrangeSky@reddit
Yea, although, if they are actually SOTA models then it's like "hoorayy! A 3 trillion parameter model just got open weighted! I hope you're ready, dear laptop! Lemme go grab the astroglide!"
carnoworky@reddit
It's that Brazzers meme, except instead of five guys it's just one guy the size of Godzilla.
Zanion@reddit
Nobody worth a damn stayed to work under Wang, so who cares. Meta AI is just an LLM feature factory now.
Plabbi@reddit
This sub has the choosiest beggars of all time.
Live-Crab3086@reddit
beggars? the hyperscalers tripled the price of personal computer equipment with their bs hype circlejerk. they're telling us they're going to replace our white-collar jobs, breaking the long-standing social contract that education leads to a good life. they're lucky people don't have torches and pitchforks anymore.
Plabbi@reddit
"Choosy beggar" is a term for people who are dependent on the goodwill of others, while simultaneously making entitled requests despite being in no position to negotiate.
dtdisapointingresult@reddit
Not only are they choosy, they're like those mentally ill aggressive beggars that aren't even satisfied when you give them something because it's not enough.
Imagine if Meta's models were more focused on language tasks than benchmaxxing corporate worker tasks. The reaction on the internet (losers on reddit and Twitter) would be overwhelmingly to dunk on Meta. This very thread is evidence. It makes it bad from a PR POV for Meta to release anything other than a benchmaxxed corpo model.
It's like when Elon released Grok 2. After like 2 months of posts on this sub gloating about how Elon is a liar and a scammer and all the other redditorisms, saying Grok 2 will never come out, then Grok 2 finally came out. Did any of those people backtrack? Did any of them express anything that came close to 'I might have been wrong?' Of course not. They transitioned directly to saying that Grok 3 will never be released because Elon is a liar and a scammer.
There is really nothing worse than an entitled redditor. They are actively sabotaging any goodwill some labs might have towards us, because they prefer to be smug redditors who "dunk" on unpopular CEOs over having good models in their hands.
Plabbi@reddit
Exactly.
People here already outraged over a simple announcement that they will receive a new model, free of charge. From a private company which has no obligation to do this.
LagOps91@reddit
are you kidding me? just a headline and you need to pay to read the rest of it? yeah i call bullshit, the website only wants to bait people into paying.
Tall-Ad-7742@reddit
i don't know how serious this website is but i just had to use a fake email and got access to the text (no money needed)
LagOps91@reddit
well okay, websites that hide their articles usually want money. and yeah... not giving away my email either.
emprahsFury@reddit
lol Axios is a well known news site.
Randomdotmath@reddit
LOL, I couldn't care less if Axios is 'well-known'—this page still a crappy rag peddling rumors mixed with blatant subjective bias, and it's paid content to boot. And you're proud of that? 😂
LagOps91@reddit
literally never heard of it
gnnr25@reddit
https://i.redd.it/emcdeut8eotg1.gif
Patq911@reddit
It's big in government/political news. If you have never heard of it you are just not that plugged in. That's ok, but it is a very well known new organization.
Technical-Earth-3254@reddit
And the 2nd paragraph starts with ai slop lmao
Cool-Chemical-5629@reddit
Why it matters it starts with AI slop? 🤭🤣
abkibaarnsit@reddit (OP)
Added the text
tsukuyomi911@reddit
Their llama models were fire. Hope the next batch is equally good.
GreenHell@reddit
Except for Llama 4, which sees less usage on Openrouter than Llama 3.1
ninjasaid13@reddit
I don't think alexandr wang has the experience.
dtdisapointingresult@reddit
You don't need to be an expert in the domain to manage a division. He's well-versed, that's enough. He's been in AI for a long time now.
Look at how Elon Musk with SpaceX and Neuralink. He didn't need to be a rocket engineer or a brain surgeon. He just needed to hire and manage the right people.
Not saying Wang's gonna be great. I just hope we get some open models out of it.
TheRealMasonMac@reddit
Isn't Elon Musk basically useless in day-to-day operations? There's no way he's actually running a billion companies.
dtdisapointingresult@reddit
Yeah I think he just decides who to put in charge of each thing, and checks on their progress. Like a senior dev vibecoding.
__some__guy@reddit
I wonder if they can still release something competitive.
Their last models (Llama 4 series) were so irrelevant, I couldn't remember them without asking Perplexity.
Mountain_Patience231@reddit
open weight or open soruce
Zanion@reddit
Nobody worth a damn stayed to work under Wang, so who cares.
ConfusedLisitsa@reddit
Metà models have not been relevant for more than two years
abkibaarnsit@reddit (OP)
breadfruitcore@reddit
This really can't be understated, many out-of-the-loop grandpas and grandmas are heavily exposed to AI via these platforms. People talk about Apple and Google integrations but the shitty Llama models you can chat directly in WhatsApp nobody talks about has been embedded insanely deeply for 1-2 years now. And WhatsApp E2EE won't even matter here because they can just add a different terms of service for their chatbot.
If Meta's AI strategy is to add another data-devouring channel to their arsenal instead of going after high value-added use cases, then opening their model doesn't cost them anything.
Dany0@reddit
Bro is younger than me 💀💀💀
Hot_Turnip_3309@reddit
if they are announcing an announcement, then that means the models are going to suck and they are depending on hype. So you can save your time and skip them when they come out.
CheatCodesOfLife@reddit
I'll probably end up trying them for the sake of nostalgia
kiwibonga@reddit
I heard Winamp really whips it.
ThunderWriterr@reddit
No, thank you!
I'm sure you are a wonderful model, but you're not really my type.
TopChard1274@reddit
Of course they do; they’re reported to be crap
silenceimpaired@reddit
Congrats all they released 1b model and a 1T model (based on Deepseek architecture) - probably someone in the future.
What I wouldn’t give for a good 70b model or an equivalent MoE.
ttkciar@reddit
You might want to check out K2-V2-Instruct, by LLM360. It's a fully open-source 72B dense trained from scratch using augmented datasets (available on HF) and plain-jane llama architecture (so jfw on all major inference stacks).
I've been really impressed by its cleverness and long-context competence.
silenceimpaired@reddit
I like it for sure. Not idea for creative writing editing, etc
Limp_Classroom_2645@reddit
nobody cares about announcements of an announcement
either release it or stfu
the_bollo@reddit
"Here, YOU throw this away."
EbbNorth7735@reddit
Oh come on guys, it might be really good and more competition is always welcome. Worst case we don't use it but ideally we give them valid feedback and they keep working on it until they product something competitive. META has released a lot of really good and solid AI models in other areas such as SEM and translation.
LelouchZer12@reddit
Could this be dinov4 ?
mindwip@reddit
What's there definition of large models.
Of they release 400b and I see models ok great. Keep your 1t models
If there saying 120b is large ok by.
You can't claim safety from advanced models and then say o but ours are not sota....
Creepy-Bell-4527@reddit
I don't mean to sound ungrateful, but...
ok
sleepingsysadmin@reddit
avocado was rumoured awhile back, should have been out already.
Working_Sundae@reddit
It's performance was reportedly worse than Gemini 2.5 Pro in unnamed benchmarks so they paused the release and continued more post training
Mochila-Mochila@reddit
kbyethx
Few_Painter_5588@reddit
That's good, more open weight models the better.
Syphari@reddit
This is like someone giving you their old run down bicycle that’s all messed up and the chain is crappy and rusted but they expect you to be excited about getting a useless jacked up bike that’s no one else wants lol
Lissanro@reddit
Full article below. OP, in the future please copy actual text in your post or link to better sources that are easily accessible.
Scoop: Meta to open source versions of its next AI models
Meta is preparing to release the first new AI models developed under Alexandr Wang, with plans to eventually offer versions of those models via an open source license, Axios has learned.
Why it matters: Meta has been the largest U.S. player to let others modify its frontier models, and there has been growing speculation the company might retreat from that strategy altogether.
Before openly releasing versions of the new models, Meta wants to keep some pieces proprietary and to ensure they don't add new levels of safety risk, according to sources.
Between the lines: The move fits with Wang's view that Meta can be a force for democratizing access to the latest AI technology and ensuring that there is a U.S.-made option that is open for developers.
Wang sees Anthropic and OpenAI as increasingly focused on delivering their models to governments and the enterprise. By contrast, Meta's effort is focused on consumers, per sources. Meta wants its models distributed as widely and as broadly as possible around the world.
The big picture: Meta has said the first family of models is designed to help it catch up to rivals after its last Llama 4 family fell significantly behind, with an aim that future models that can lead the industry.
Yes, but: The leaders aren't standing still. Both OpenAI and Anthropic are hinting that their next models, also expected to drop soon, represent significant advances.
Meta knows its new models may not be competitive across the board with the coming ones from those labs, but believes it will have areas of strength that appeal to consumers, the sources said.
And don't expect a full return to Meta's earlier openness. Wang has indicated that some of its largest new models will remain proprietary — a shift toward a more hybrid strategy, according to sources.
Meta argues it still reaches users more broadly than rivals by embedding AI into WhatsApp, Facebook and Instagram — free services with global scale that competitors can't easily match.
Our thought bubble: Meta's approach increasingly looks like a hedge: open enough to win developer mindshare and shape the ecosystem, but closed where it believes the biggest models confer a competitive edge.
That mirrors a broader industry shift, where even companies that champion openness are pulling back on their most powerful systems. Alibaba recently kept its most powerful new Qwen models proprietary, reversing its own open-source playbook.
Context: Wang joined Meta last year as part of a $15 billion deal with Scale AI, where he was CEO.
johnfkngzoidberg@reddit
So? These “may happen” hype lowkey advertising posts are getting out of control. Come back when I can download some weights.