Pinned Tweet
So happy I can finally share this with you all.
We spent months building AIRA₃, entered it in a live Kaggle competition against ~4,000 human teams, and watched it teach an LLM to reason on its own.
Gold 🏅 First ever by an autonomous AI system.
Unhackable eval. Early RSI ♻️
As a test of our progress to advance the frontier of AI research, in June we entered the next generation of our autonomous AI research system, AIRA₃, in a live Kaggle competition run by NVIDIA to fine-tune a 30B Nemotron model. The challenge was to teach the model to reason better — all competitors had access to the same information and were graded externally on a private test set.
AIRA₃ placed 8th out of ~4,000 teams to win Gold, outperforming human competitors who had access to the same frontier tools.
We believe this is a reliable signal that AIRA₃ can improve a targeted capability of an AI model at a level similar to human experts.
Karen Hambardzumyan retweeted
Today we published a technical report on Muse Glimmer, our 30B open-weight model built for agentic tasks on consumer hardware.
It runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision, and leads comparably-sized open models on tool use and multi-step planning.
Read about the model architecture, distillation recipe, quantization and more: go.meta.me/4Bcvxol
Karen Hambardzumyan retweeted
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.
Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.
Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token context window
→ Natively multimodal across text, image, video, audio
Karen Hambardzumyan retweeted
Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.
Happy to share that AIRA₂ has been accepted at NeurIPS 2026! 🎉
Sequel to AIRA₁ (NeurIPS 2025 Spotlight), and predecessor to AIRA₃ -- which just won its first live Kaggle gold. Great to see this work compound.
"It's like the IMO and IOI results from OpenAI a year ago, but this time for RSI."
Given how fast the meta team was able to approach the lagging frontier, I find it compelling that this model scored so well.
It's like the imo and ioi results from OpenAI a year ago, but this time for RSI.
We see where the frontier is for math, so I think this focus bodes well
15
882
@MartinJosifoski @EdanToledo @NickyBaldwin3 @RishiHazra95 @MichaelKuchnik absolute beasts, all of them.
Three generations of AIRA and this crew never once stopped cooking 🍳 Genuinely lucky I got to build it with them, and everyone at MSL who helped!
Stay tuned for more ⚡
5
233
Thrilled to see the work on AI Research Preference Models (RPMs) finally shared.
Amazing work lead by @basselalomari_ , @_tomwithanh @tingchenfu and the team. 🧠
AI research agents can generate hundreds of ideas in seconds, but evaluating each can take days of GPU time. When compute is limited, which idea deserves execution?
We introduce AI Research Preference Models (RPMs) to assess ideas & focus compute on the most promising paths 🧵👇
17
470
Karen Hambardzumyan retweeted
Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.
Karen Hambardzumyan retweeted
Meta's chief ai officer @alexandr_wang :
"Internally at Meta, we have seen cases where, if you develop the right agentic loop and have the right evaluation system and metric for the agents to optimize, a swarm of agents can accomplish more than a team of 100 engineers. They can do it very handily, actually, very easily."
Alexandr Wang was talking to CEO of Y Combinator @garrytan at Y Combinator Startup School 2026
----
full video link in comment
Karen Hambardzumyan retweeted
We pointed Muse Spark 1.2 at a kernel optimization task and let it run. 1,000+ tool calls over 24 hours on NVIDIA Hopper. It kept finding substantial improvements well beyond the initial exploration phase.
Karen Hambardzumyan retweeted
Replying to @MartinJosifoski
Great to see this research out in the world, great work @MartinJosifoski and team!
Our next-gen AI Research Agents are here. 🚀 Building AIRA₂ gave us incredible insights into solving bottlenecks, and we're excited to share them. Async evolutionary exploration, parallel agents, closed generalization gap, scaling laws, aha moments, and more.
Thread & paper👇
Excited to share AIRA₂ — our next-generation AI Research Agents for ML that address key bottlenecks to scaling.
AIRA₂ achieves SoTA on real-world ML tasks from MLE-bench-30 (81.5% vs 72.7%), exceeds human SoTA on 6/20 diverse AI research tasks from AIRS-Bench (and hacks another 5), while exhibiting strong, predictable scaling properties.
To push the frontier of AI Research, we need systems that scale well. Developing AIRA₂, we learned a lot about the bottlenecks and what it takes to resolve them — insights already driving our next iteration:
1/
Karen Hambardzumyan retweeted
Introducing - AIRS Bench, a benchmark for “AI Researcher Agent”. Agents attempt 20 open ML problems starting from zero code (full research loop). And yes, they beat SOTA in few cases (read more below!) arxiv.org/abs/2602.06855
We are entering the era of agentic science. But what does it take to build and scale these systems reliably? 🤔
I'm not there to geek out with you, but my co-authors @EdanToledo, @MartinJosifoski and @RishiHazra95 are at #NeurIPS2025 and eager to chat - Don't miss our poster!
Come see our poster today! If you’re interested in research agents and how to unlock scaling for them, come for a chat!
2
325
Karen Hambardzumyan retweeted
New @AIatMeta paper shows AI agents do better when their first ideas are diverse, not similar.
The authors studied 11,000 agent runs on Kaggle tasks and measured diversity in the first 5 plans.
Here diversity means trying model families, not tweaks of one model.
Agents whose first ideas covered many types scored higher across many agent setups.
When the system prompt was changed to push similar ideas, benchmark scores fell by 7 to 8 points and valid submissions dropped.
This suggests that having several distinct plans gives the agent more chances to find a solution it can build.
Scaffold settings like sibling memory, staged complexity hints, and explicit variety requests raise diversity, while temperature changes do little and better agents spend more time on successful code.
----
Paper – arxiv. org/abs/2511.15593
Paper Title: "What Does It Take to Be a Good AI Research Agent? Studying the Role of Ideation Diversity"
Karen Hambardzumyan retweeted
1/ Introducing Isaac 0.1 — our first perceptive-language model. 2B params, open weights. Matches or beats models significantly larger on core perception. We are pushing the efficient frontier for physical AI.
perceptron.inc/blog/introduc…
Karen Hambardzumyan retweeted
been waiting for this for a while. really excited for the first release from this amazing team
1/ Introducing Isaac 0.1 — our first perceptive-language model. 2B params, open weights. Matches or beats models significantly larger on core perception. We are pushing the efficient frontier for physical AI.
perceptron.inc/blog/introduc…
Karen Hambardzumyan retweeted
We discovered alien intelligence in sand, and can now play its dreams in real time with a mouse and keyboard.
Congrats to the team!
Now, can it run Doom? 🤔
Karen Hambardzumyan retweeted
AIRA strikes again! This time we conduct an in-depth study of research agents on MLE-Bench (i.e. kaggle competitions). We find that while exploration and search matter, the biggest delta is due to our more robust software stack. We are open-sourcing all of this to allow YOU to push this work further.
AI Research Agents are becoming proficient at machine learning tasks, but how can we help them search the space of candidate solutions and codebases? Read our new paper looking at MLE-Bench: arxiv.org/pdf/2507.02554
#LLM #Agents #MLEBench



























