🔥Telegram İfşa

Check out the generative vision related release too meta.ai/?icebreaker=im… Imagine Flash generates the image as you type You can also "Animate" your images! (technique based on Emu Video emu-video.metademolab.com) Kudos to the team for putting this out :)

If you want to dive into the details of Imagine Flash, check out the research paper that we published on this today. Really proud of how quickly we progressed from research to product with Imagine Flash: ai.meta.com/research/publi…

While there is a lot to do in benchmarking progress it is still great to see Llama 3 models trending well. Research is iterative and groundbreaking at the same time. Evaluation of progress at the two ends of spectrum is fraught with debates. Thank you Lmsys!

An awesome memory from when I was actually doing some "real" (coding) work. Now, working on Llamas to do some of the coding :-)

Excited by the progress the teams have made in both 3D Generation and texturing the generated models especially texturing videos at 10x faster. More details in the papers: Meta 3D AssetGen paper ➡️ go.fb.me/87tktg Meta 3D TextureGen paper ➡️ go.fb.me/tvbdf8

I am excited to have Russ, Daniel and Jing Yu join the Llama team to work on Multimodal agents. We have to tackle many open research problems to enable agentic capabilities. Looking forward to working together. Welcome to the Llama family

Super excited to have you on the Llama team, the journey to push frontiers of intelligence and bringing that to everyone is still very early and embarking on that together is a great opportunity!

Meta Movie Gen is just freakin cool! Generative Video Foundation models with this quality, precise editing and personalization unlock value for creators, new creative tools and enable Agents that can interact in richer ways closing the loop on learning to unlock world models!

The Audio generations conditioned on video are really awesome and get us closer to a true multimodal model with a unified understanding of the world. Generation & Understanding can seem more unrelated at surface but will be much more interconnected as we iterate!

Last release for Llama 3 family before we bring Llama 4 to the world in 2025! Excited that folks can reap the benefits of 405B level performance at 70B cost with this update.

Excited that you will join the llama journey and look forward to making faster progress for bigger and better llamas!

The capability & efficiency cycles in AI always followed each other. The debates on research vs engineering, which country or model is winning will rage on, it is heartening to see AI progress continuing to happen at a faster rate thanks to everyone pushing (hopefully together!)

Conferences are a great way to come together and learn in a fast moving world. Tune in!!

Great to see the open weight movement picking momentum and everyone is leaning towards! We should compete and collaborate towards a joint human<->agent intelligence future!

As of today, Llama 4 Maverick offers a best-in-class performance to cost ratio with an experimental chat version scoring ELO of 1417 on LMArena. It's wild to think Llama was a research project a couple of years ago & amazing to see how much progress we've made in the last two

BREAKING: Meta's Llama 4 Maverick just hit #2 overall - becoming the 4th org to break 1400+ on Arena!🔥 Highlights: - #1 open model, surpassing DeepSeek - Tied #1 in Hard Prompts, Coding, Math, Creative Writing - Huge leap over Llama 3 405B: 1268 → 1417 - #5 under style control

We lived in the era of "Data + Flops => Model" and we are now seeing the era of "Model + Flops => Data" and the loop between the two will continue the current AI progress for a while until we hit new walls...

Tune in for our first edition of Llama developer conference!

Congratulations to Andrea and team!