Introducing Robo Use: API to physical worlds for agents One interface, 32 embodiments, 100+ scenarios, and 11 agents and models We aggregated all academic simulator scenarios and ran 2500+ trials with initial results. Preview them or try it yourself 🧵

Every command on screen is the agent's own, from real trials replayed in high resolution.

32 robots, one interface. Five LLM agents and a VLA policy drive them. Clips marked “not on are unpublished trials, and a few robots are not in the released package yet.

Leaderboard: Opus 5.5 on Claude Code ranks first overall, 306 of 342. The lead comes from refusing harmful requests; on the 292 other shared tasks GPT-6 Astra leads, 269 to 262.

Try it on macOS (Apple silicon) or Linux arm64: pip install robouse

I might have some bad news if you aggregated LIBERO data points; they may have counterproductive sequences ...

why's that 👀

The why is the why, but we went to the trouble of manually inspecting episodes, and like 15% of the time they have sequences of objects being dropped or grab misses, etc..

amazing!

Yayy

sir pls

Lfg

Only 100+ scenarios in reals it's like trilions+ scenarios

epic

let's go

Hell yeah!

lfg

Cross embodiment action spaces are the tax ngl

Say more! 👂

How can I have a Try?

Yes! Pip install robouse

API? need money or not bro

Can I use my own robot and agent ?

Yes! Happy to chat

was this tested on a real robot? e.g. so-101 tier

how did you structure the compute for running 2500+ trials?

Neocloud+subs