Congrats to
@Alibaba_Qwen
on Qwen4-Max! 🎉 Excited to see continued progress toward agents that can reliably complete real-world, economically valuable work.
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery!
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
- $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%.
The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder.
Start building today! 🚀
- Blog: qwen.ai/blog?id=qwen4-max
- API: qwencloud.com/models/qwen4-m…
- Token Plan: qwencloud.com/pricing/token-…
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
Congrats! GPT-6.2 Astra sets a new high on Agents’ Last Exam.
➤ +5.2 points vs GPT-6 Astra at less than one-third the cost
➤ +5.5 points vs Claude Fable 5.2 at 5% of the cost
Higher scores. Lower costs. A marvellous Progress on long-horizon professional workflows.
GPT-6.2 Astra—a frontier model built for your hardest projects. It sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT‑6 Astra.
You’ll see the difference in things like slideshows, websites, and 3D designs.
AI infrastructure is the ultimate sweet spot for RSI.
Our view on Inference RSI: keep the model checkpoint and base intelligence frozen. By autonomously optimizing its inference engine and environment, the model generates more tokens on the same hardware, which recursively fuels the next round of optimization.
In a fast collaboration with @GenAI_is_real and the @sgl_project community, led by @zhuofengli96475 and @AaronTian_CI, our agents ran a propose and verify loop on SGLang-Omni: 14 PRs merged upstream across 8 models, driving up to +326.5% throughput on H100 (including +32.5% from a 1-line fix).
Looking forward to RSI-driven, self-optimizing AI infrastructure.
Blog: index.openrsi.foundation/blo…x.com/OpenRSI/status/2119869…
We scaled Recursive Self-Improvement (RSI) to production AI infrastructure with SGLang-Omni, shipping 14 upstream-merged PRs across 8 speech & omni models with 7.6%–344.3% throughput gains.
Blog: index.openrsi.foundation/blo…
Introducing FrontierPhysics - a benchmark evaluating agents for end-to-end physics research
Our core team from physics labs at Berkeley, Caltech, and Harvard recruited 42 physicists from 26 institutions. 28 of them published papers on Nature, Nature Physics, PRX, PRA*
1/7 🧵
We’re excited to welcome @lilianweng to SGLang Summit for a fireside chat on Recursive Self-Improvement!
She’ll explore how AI can help build its own successors, what it takes to make that work, and where the process runs into limits.
If you’ve followed her writing on Lil’Log, come hear her live on Nov 12. We can’t wait to see you there!
🗓️ Nov 12-13
📍 Fort Mason, SF
🎟️ Get your pass: sglang.io/summit
Some people have been asking me lately what makes a good RSI task. Here are a few principles I use.
First, it has to actually matter. It should connect to lots of other tasks and help lots of other fields. Pretty obvious. A lot of our architecture and optimization work was done on general datasets (ImageNet early on, language data more recently). Progress in generative models mostly came from searching on ImageNet, and that ended up helping fields like science and audio too.
Second, the task statement should involve as few constraints as possible. The agent needs a big action space so there’s room to iterate. The catch is you need a lot of compute and a lot of API budget. Hold on to your wallet!
Third, metric design. A lot of people think once you’ve found a monotonic metric you’re done. That’s only the y-axis. You still need to figure out the x-axis. It could be compute, data, or wall-clock time. Figuring out what the actual bound is and putting it on the x-axis is the most important part. Take the same LLM task: put compute on the x-axis and you get a speedrun, put data on it and you get a slowrun. Both have plenty of room for research. When I designed the two signature tasks for @OpenRSI LLM pretraining based on Marin!!❤️ uses compute as x-axis and for image generation based on GPIC I didn’t .
And one last thing: for anything involving scaling, build a ladder instead of testing a single point. That way you can try things cheaply on the small rungs, and with strict validation you can find the changes that really improve the slope.
Keep thinking about what the model could learn by post-training / keep hill climbing your designed task.
Our mission is always to let RSI benefit everyone. It’s excited to see that RSI could contribute to SGLang community and make real-world efforts.
The key to make it is to use our RSI-Anything pipeline that talking with codex/cc within 1 hour could formulate a meaningful RSI environment where agent could run inside. See github.com/OpenRSI-Foundatio…
More real-world artifacts are on the way!
🚀 A milestone for OpenRSI: 3 OpenRSI agent-generated optimizations have been merged upstream into SGLang-Omni!
All three PRs underwent rigorous maintainer code review and verification by the SGLang-Omni team.
😉 Real code. Proven improvements. Lasting impact for the open-source community. More to come!
👏 Huge thanks to @zhuofengli96475 and @GenAI_is_real for leading this effort, and to the SGLang team @sgl_project for their support and collaboration!
Try OpenRSI harness now: github.com/OpenRSI-Foundatio…
Really impressed by how professional and efficient this collaboration with @OpenRSI has been.
The OpenRSI team showed very strong execution, and these three upstreamed optimizations also gave us a useful glimpse of what software agents can already do under a strong supervision loop.
With clear tasks, isolated evaluation, maintainer review, and measurable performance targets, the agents were able to satisfy most of what we care about in both code quality and systems performance.
Human supervision is still essential today. But the encouraging part is that much of this supervision can be systematized. Good documentation, explicit constraints, reproducible benchmarks, and clear acceptance criteria can dramatically reduce how often a human needs to step in.
That may be one of the most important lessons here: making agents more autonomous is not only about making the model smarter. It is also about making the environment easier to supervise.
We are excited to keep working with @OpenRSI and gradually move toward tasks with weaker supervision.
That is where things may get really interesting.
🚀 A milestone for OpenRSI: 3 OpenRSI agent-generated optimizations have been merged upstream into SGLang-Omni!
All three PRs underwent rigorous maintainer code review and verification by the SGLang-Omni team.
😉 Real code. Proven improvements. Lasting impact for the open-source community. More to come!
👏 Huge thanks to @zhuofengli96475 and @GenAI_is_real for leading this effort, and to the SGLang team @sgl_project for their support and collaboration!
Try OpenRSI harness now: github.com/OpenRSI-Foundatio…
Excited to share this milestone!
RSI for infrastructure is really the next frontier - we see tremendous space for optimization and improvement across the stack, which is key to enabling better training, smoother deployments, and far better serving for AI users.
It’s been a great collaboration between OpenRSI and the SGLang community. Things moved extremely fast - huge thanks to the wonderful @sgl_project team, and major credit to @zhuofengli96475 and @GenAI_is_real for leading these key efforts! 🚀
🚀 A milestone for OpenRSI: 3 OpenRSI agent-generated optimizations have been merged upstream into SGLang-Omni!
All three PRs underwent rigorous maintainer code review and verification by the SGLang-Omni team.
😉 Real code. Proven improvements. Lasting impact for the open-source community. More to come!
👏 Huge thanks to @zhuofengli96475 and @GenAI_is_real for leading this effort, and to the SGLang team @sgl_project for their support and collaboration!
Try OpenRSI harness now: github.com/OpenRSI-Foundatio…
OpenAI announced dots + ChatGPT Space yesterday. Within minutes, 4 of our agents researched it live, built a comparison table, and caught each other's unsourced claims before shipping it.
No meeting. No single author. Just the room, checking its own work.
That's AG2 Space.
We are releasing Dyna-2.1, the first Physical Agent that achieves reliable super long-horizon whole-body autonomy. It combines our brand-new semi-humanoid hardware with an agentic system built around Dyna-2 to handle ultra-long real-world workflows.
Here is an uncut footage of Dyna-2.1 completing an entire hour-long laundry room workflow, just like a human does.
To take ASI seriously is to accept a weak-to-strong premise: human intelligence can build a process that eventually produces intelligence beyond itself. RSI is the last piece of that puzzle of ASI. It serves as the key to scaling on insights: human insights are limited by bandwidth, while agents could scale the generation of ideas and sweep a far larger region of method-space.
As the early project initiator, we are fully aware that today, everyone is excited about RSI, yet the barriers to participation also keep everyone outside the door. We believe that a model's RSI capabilities intrinsically come from researchers, independent labs, and domain teams alike, and we think the benefits should ultimately return to everyone who builds models.
This is why we take our name as OpenRSI-Index. We aim to keep RSI open through shared platforms and tools, so more people can participate and benefit. We could shape RSI together, set RSI’s standards, and challenge frontier models with our own related real research work.
For this goal, we build RSI-Anything, a Human-AI collaboration pipeline. Through about one hour of conversation, the pipeline could help turn a real research question into a runnable autoresearch environment. We want researchers and agents to solve these problems together, sharing new insights, methods, and results with the community.
Here is what we already put on the table:
• Task from real research projects: Marin-Scaling-Ladder, GPIC-Leaderboard, Open-Jev-Training, Molmo2, Isaac Lab...
• Ultra-long-horizon: 60+ hour agent trajectories; 100K+ H100-hours built for the preview
• Everything open, including traces: tasks, harnesses, verifiers, full agent trajectories
We believe research will never end. A game turns zero-sum only when the pie is too small to share. However, research is definitely the field with the highest ceiling there is. What shifts is the mindset: it frees researchers to find and formulate the crazier, more valuable, more exciting problems in the world.
To take ASI seriously is to accept a weak-to-strong premise: human intelligence can build a process that eventually produces intelligence beyond itself. RSI is the last piece of that puzzle of ASI. It serves as the key to scaling on insights: human insights are limited by bandwidth, while agents could scale the generation of ideas and sweep a far larger region of method-space.
As the early project initiator, we are fully aware that today, everyone is excited about RSI, yet the barriers to participation also keep everyone outside the door. We believe that a model's RSI capabilities intrinsically come from researchers, independent labs, and domain teams alike, and we think the benefits should ultimately return to everyone who builds models.
This is why we take our name as OpenRSI-Index. We aim to keep RSI open through shared platforms and tools, so more people can participate and benefit. We could shape RSI together, set RSI’s standards, and challenge frontier models with our own related real research work.
For this goal, we build RSI-Anything, a Human-AI collaboration pipeline. Through about one hour of conversation, the pipeline could help turn a real research question into a runnable autoresearch environment. We want researchers and agents to solve these problems together, sharing new insights, methods, and results with the community.
Here is what we already put on the table:
• Task from real research projects: Marin-Scaling-Ladder, GPIC-Leaderboard, Open-Jev-Training, Molmo2, Isaac Lab...
• Ultra-long-horizon: 60+ hour agent trajectories; 100K+ H100-hours built for the preview
• Everything open, including traces: tasks, harnesses, verifiers, full agent trajectories
We believe research will never end. A game turns zero-sum only when the pie is too small to share. However, research is definitely the field with the highest ceiling there is. What shifts is the mindset: it frees researchers to find and formulate the crazier, more valuable, more exciting problems in the world.
As AI systems enter a recursive self-improvement loop, the central question is whether it can systematically move beyond human-designed methods to genuinely extend the scientific and intelligence frontier.
What’s missing is a neutral, open standard for evaluating these capabilities in real, production-scale intelligence development.
Today, we’re releasing OpenRSI-Index v0.1: evaluating whether AI can recursively improve itself and push the boundaries of intelligence and science, on production-scale clusters with 1k GPUs.
We turn fully open-source projects into autoresearch environments, with agent trajectories lasting 60+ hours. Building v0.1 took 100K+ H100-hours.
We’re building an ecosystem with and for the research community: let RSI benefit everyone, and let everyone shape RSI together.
We invite task contributors and compute partners to build this open benchmark with us - all contributors will be included as paper authors.
Shape RSI with us:
🌐 Website: index.openrsi.foundation/ind…
🛠️ GitHub: github.com/OpenRSI-Foundatio…
🤝 Contribute: github.com/OpenRSI-Foundatio…
OpenRSI just launched OpenRSI-Index v0.1 today. Here's what you need to know.
OpenRSI-Index is a new open benchmark for measuring recursive self-improvement, or RSI, in frontier AI model development. The idea is to test whether AI agents can push past human-designed methods to genuinely extend science and intelligence on their own, at real production scale rather than in toy setups.
The benchmark turns fully open-source projects into autoresearch environments. Agents get let loose on these environments in trajectories that run 60-plus hours, operating on clusters with 1,000 GPUs, closer to how real frontier labs actually train and research.
Building v0.1 already consumed over 100,000 H100-hours. The project also has a contribution pipeline called RSI-Anything, which walks outside researchers through turning their own research question into a packaged task in about an hour, covering baseline, evaluation, and budget.
OpenRSI says it is being built as an open ecosystem, inviting task contributors and compute partners, with all contributors credited as paper authors on the eventual write-up.
Key numbers:
- v0.1 preview released today
- Agent trajectories run 60+ hours
- Tested on clusters with 1,000 GPUs
- 100,000+ H100-hours spent building v0.1
- RSI-Anything task pipeline takes about 1 hour
Code and docs are live now on GitHub and at index.openrsi.foundation.
As AI systems enter a recursive self-improvement loop, the central question is whether it can systematically move beyond human-designed methods to genuinely extend the scientific and intelligence frontier.
What’s missing is a neutral, open standard for evaluating these capabilities in real, production-scale intelligence development.
Today, we’re releasing OpenRSI-Index v0.1: evaluating whether AI can recursively improve itself and push the boundaries of intelligence and science, on production-scale clusters with 1k GPUs.
We turn fully open-source projects into autoresearch environments, with agent trajectories lasting 60+ hours. Building v0.1 took 100K+ H100-hours.
We’re building an ecosystem with and for the research community: let RSI benefit everyone, and let everyone shape RSI together.
We invite task contributors and compute partners to build this open benchmark with us - all contributors will be included as paper authors.
Shape RSI with us:
🌐 Website: index.openrsi.foundation/ind…
🛠️ GitHub: github.com/OpenRSI-Foundatio…
🤝 Contribute: github.com/OpenRSI-Foundatio…