Joined January 2009
- Tweets802
- Following898
- Followers1.7K
- Likes4.2K
Excited to see so many of my ex-colleagues come together to solve this important problem for society.
Congrats on the raise @manohar_paluri @kevinweil @dpatil - proud to play a small role in this journey!
warp.xyz
💥 Introducing Warp Sciences, a new company applying AI and robotics to help the world develop new medicines faster.
We’ve raised a $215M Series A led by @lightspeedvp to bring frontier AI out of the browser and into the lab, and build scientific capacity here in America.
I co-founded Warp with my long-time friend @dpatil, FAIR co-founder @manohar_paluri, and drug hunter Sol Reisberg. The early team is full of incredible people like @bkspears9, who helped achieve Q>1 fusion at LLNL.
8 years, 3 AI orgs (Facebook AI, Gen AI, MSL), 6 foundation models, and countless boba-fueled war rooms later, I wrapped up my time at @AIatMeta.
Getting to train models that helped billions of users has been an incredible and humbling ride. From the early days of scaling face recognition and misinformation detection, to early on-device models for glasses, to releasing new image & video generation models, to open-sourcing several versions of Llama, and most recently, Muse Spark showing up in Meta AI - any one of those experiences would have been the highlight of a career. And here I am lucky to have been part of all of them.
To all my friends at Meta who trusted and built with me through several intense transformations: thank you. Take care and I’m always a message away.
What’s next for me? I’m going to continue training new foundation models, but this time for what I believe is the next frontier: Physical AI. More on that soon.
Back to building! 🚗
Roshan Sumbaly retweeted
Following-up: Muse Spark is 100% legit. It handles my general queries as well as, if not better than, other frontier models. And echoing @borrowed_ideas: 'contemplating mode' is a true leading-edge feature.
Congrats to @alexandr_wang and the entire MSL team! Can't wait for more.
I tested $META's Muse Spark over the last few hours and came away net positive. 3 main takeaways:
1) Quality: It's a very good model. Not quite frontier but good. It showed comparable performance vs Opus 4.6 across web data search, PDF parsing, and general knowledge/conversation/writing. It's worse at coding. Both models solved an easy coding task, but Muse Spark failed the hard one while Opus one-shotted. The image-gen is also worse than ChatGPT. But all in, it's a legitimate and usable general model. Lots of room to develop the UI further (ie it should show a map when recommending local restaurants) but the underlying model itself is impressive.
2) Speed: Notably, Muse Spark answered almost instantly while Opus 4.6 felt borderline unusable at times. I'm a huge Anthropic fan but latency has become a major issue. Simple answers take too long and multi-step agent flows break more often. Meta seems to have more available compute which is a real factor going forward.
3) Scaling: Meta hasn't published a full model card so we're working with limited disclosure. But the graphs below might be the most important part of the release. Their rearchitected pretraining stack shows a near-linear relationship between RL compute and accuracy. If that holds, Meta has a clear path to training much larger, more intelligent models. That's arguably more consequential than Muse Spark itself.
All in, it's positive. Muse Spark is a good, usable model, it's being served smoothly, and Meta looks to be on an encouraging trajectory.
Roshan Sumbaly retweeted
okay this is too exciting :)
meta AI is now #2 in the app store, top AI app!
we are so back!
Roshan Sumbaly retweeted
I gave the new Meta Muse Spark model a bunch of assorted assets and it was able to carefully extract them and place it into a functioning app.
I've never seen something with a clear intuition for design and product-first thinking.
This is the first AI that *gets* UI design.
Roshan Sumbaly retweeted
Meta is back in the Arena!
Muse Spark debuts as a top frontier model across both Text and Vision:
- Text Arena: #3 tied with Gemini-3.1-Pro and Claude-Opus-4.6
- Vision Arena: #2 tied with Claude-Opus-4.6
This marks Meta’s first major release since early 2025.
Highlights:
- #4 Hard Prompts, #6 Coding, #9 Creative Writing, #10 Instruction Following, #27 Expert
- #3 tied for Business, Management, & Financial Ops, #7 Legal & Government, #12 Writing & Literature
Meta is back at the frontier. Huge congrats to @AIatMeta on this incredible milestone!
Introducing Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs.
Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.
Muse Spark is available today at meta.ai and the Meta AI app. We’re also making it available in private preview via API to select partners, and we hope to open-source future versions of the model.
Learn more: go.meta.me/43ea00
Roshan Sumbaly retweeted
Meta is back! Not only is it a high scoring models on AA third party eval, it is also one of the most token efficient.
Meta is back! Muse Spark scores 52 on the Artificial Analysis Intelligence Index, behind only Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Muse Spark is the first new release since Llama 4 in April 2025 and also Meta's first release that is not open weights
Muse Spark is a new model from @Meta evaluated on Artificial Analysis. We were given early access by Meta to independently benchmark the model. It is the first frontier-class model from Meta since Llama 4 Maverick was released in April 2025, and notably the first @AIatMeta model that is not being released as open weights. The release follows Meta's reorganization of its AI efforts under Meta Superintelligence Labs, and signals that Meta is re-entering the frontier race after roughly a year of relative quiet.
For context, Llama 4 Maverick and Scout scored 18 and 13 respectively on the Artificial Analysis Intelligence Index as non-reasoning models at the time of their release, while Muse Spark scores 52. Muse Spark essentially closes the gap between to the frontier in a single release.
The model is not open source and is not yet accessible via an API but Meta has shared they expect this to come soon. Meta is also integrating Muse Spark into their first party products including their Meta AI chat product, Facebook, Instagram and Threads.
Key takeaways from our benchmarks:
➤ Muse Spark scores 52 on the Artificial Analysis Intelligence Index, placing it within the top 5 models we have benchmarked. It sits ahead of Claude Sonnet 4.6, GLM-5.1, MiniMax-M2.7, Grok 4.20 and behind Gemini 3.1 Pro Preview, GPT-5.4 and Claude Opus 4.6
➤ Muse Spark is notably token efficient for its intelligence level. It used 58M output tokens to run the Intelligence Index, comparable to Gemini 3.1 Pro Preview (57M) and notably lower than Claude Opus 4.6 (Adaptive Reasoning, max effort, 157M), GPT-5.4 (xhigh, 120M) and GLM-5 (110M)
➤ Muse Spark is the second-most capable vision model we have benchmarked. It scores 80.5% on MMMU-Pro, behind only Gemini 3.1 Pro Preview (82.4%)
➤ Muse Spark performs strongly on reasoning and instruction-following evaluations. It scores 39.9% on HLE, trailing only Gemini 3.1 Pro Preview (44.7%) and GPT-5.4 (xhigh, 41.6%). The model also achieved 5th highest in CritPT with a score of 11%, an eval that is focused on difficult physics research questions. This is substantially above above Gemini 3 Flash (9%) and Claude 4.6 Sonnet (3%)
➤ Agentic performance does not stand out. On GDPval-AA, our evalaution focused on real world work tasks, Muse Spark scores 1427, behind both Claude Sonnet 4.6 at 1648 and GPT-5.4 at 1676, but ahead of Gemini 3.1 Pro Preview at 1320. On On TerminalBench Hard, Muse Spark trails Claude Sonnet 4.6, GPT-5.4, and Gemini 3.1 Pro. Muse Spark joins others in achieving a high τ²-Bench Telecom score of 92%
Key model details:
➤ Modalities: Multimodal including text and vision input, text output
➤ License: Proprietary, Meta's first frontier model not released as open weights
➤ Availability: No public API at the time of publishing. Meta expects to provide API access soon. Meta has started integration into their first party AI offering Meta AI and inside Facebook, Instagram, and Threads
Roshan Sumbaly retweeted
1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵
Say hi to Muse - Meta Superintelligence Lab's new foundation model series. Today we're releasing the first model Spark 🥑. Lots of goodness in this natively multimodal reasoning model that does well on several agentic and knowledge tasks. And it is available on meta.ai for you to play with today!
It has been almost a year of hunkering down and rebuilding our stack to scale - so lots more to talk about in the upcoming months. Let's go!
Blog: ai.meta.com/blog/introducing…
17 years ago, I took courses from @chrmanning and later TA’d a course he co-taught with @WittedNote at Stanford. The lessons in humility and patience I learned from him in those brief interactions as a young grad student still stay with me. His influence on the field of NLP and LLMs is immeasurable — and he remains one of the most humble human beings I’ve known.
Llama 4 is here with 4 models!🦙🦙🦙🦙
I'm back to share with the world what the team has been cooking. Today we're open-sourcing 2 state-of-the-art omni models (Scout, Maverick - including pre-trained weights), previewing a 3rd one (Behemoth) and will drop a reasoning one soon.
We've moved from dense models to MoEs, and pre-trained all of these with 30T tokens spanning modalities and over 200 languages:
🏕️ Scout: 17B active, 16 experts, 109B total params. I call this a "hobbyist" model as it can fit on a single GPU and is better than all our previous Llama 3 models. And with 10M context window you can really push the limits...
🐴 Maverick: 17B active, 128 experts, 400B total params. Our "workhorse" model that powers Meta AI assistant. And with state-of-the-art native multimodality support it, we hope the community can build some really cool applications...
🐳 Behemoth: 288B active, 16 experts, 2T total params. This teacher model was pre-trained on 32K GPUs and is still post-training - but has already been instrumental in enabling the above models. And it's already the best model in the world, outperforming GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks
We're just getting started! We have a few more releases lined up and are excited to share everything with the community (heck, even on a Saturday!). Our commitment to open-source is unwavering and we'll publish details of all our new techniques soon.
Exciting time to be at Meta and in the AI community - come cook with me as we push the limits of post-training!
Start here llama.com/llama4/
Blog ai.meta.com/blog/llama-4-mul…
Also you can try them out on meta.ai
🕷️
Early Christmas gift that we wanted to share with the community - last one for the year and a nice way to say good bye to Llama 3.
2025 will be the year of Llama 4!
Introducing Llama 3.3 – a new 70B model that delivers the performance of our 405B model but is easier & more cost-efficient to run. By leveraging the latest advancements in post-training techniques including online preference optimization, this model improves core performance at a significantly lower cost, making it even more accessible to the entire open source community 🔥
huggingface.co/meta-llama/Ll…
When we wrote the first research proposal for Movie Gen 6 months back we had two goals in mind:
🎬 First, accelerate creative expression for everyone - from production studios to the 100s of millions of content creators on Meta
📜Second, accelerate research in media generation foundation models, akin to what we've done with Llama
We're just getting started on #1 - we worked with filmmakers like @aneeshchaganty and @CaseyAffleck123 and got to learn so much about the artistic process. Super excited to share Aneesh's work with the world.
We're also open-sourcing evaluation sets with over 1K videos, that we hope will move the field forward and enable the next generation of foundation models - github.com/facebookresearch/…
A few weeks ago, i was given early access to movie gen from @Meta. My only instructions were to make something with it. As always, a very special thank you to my mom, who digitized over 100 hours of home video footage to make this story possible.
youtube.com/watch?v=4iIQHZ19…
Lights, camera, action - introducing Meta's Movie Gen!
Our latest breakthrough in AI-powered media generation, setting a new standard for immersive AI content creation.
We're also releasing a 92 page detailed report of what we learned, along with evaluation prompts that we hope pushes the community forward.
📽️Examples: youtube.com/playlist?list=PL…
📜Paper: ai.meta.com/static-resource/…
🖥️Site: ai.meta.com/research/movie-g…
