
Academic lab · Berkeley, United States
AgenticaBerkeley Sky Lab
A Berkeley Sky Computing Lab initiative that publishes open reinforcement learning recipes, models and the rLLM framework for training language agents.
Berkeley, United States
- Type
- Academic lab
- Based in
- Berkeley, United States
- Featured projects
- 6
About
Agentica is an open-source effort from the Sky Computing Lab at UC Berkeley. It aims to make reinforcement learning techniques for language models and agents widely available, with core contributors Michael Luo and Sijun Tan and a group of student contributors.
The project released DeepScaleR in February 2025, a 1.5B model trained with distributed reinforcement learning for math reasoning. In July 2025 it published DeepCoder, a 14B code reasoning model, the rLLM framework for training agents with reinforcement learning, and DeepSWE, a 32B software engineering agent.
Its advisors are Li Erran Li, Raluca Ada Popa and Ion Stoica. The project site notes that Agentica is becoming a startup named Peano AI.
Mission and ambitionsAgentica states that it wants to democratize reinforcement learning for language models and to build scalable systems for LLMs and agents.
Known for
- 01
DeepScaleR reinforcement-learning math model
- 02
DeepCoder code reasoning model
- 03
The rLLM agent training framework
- 04
DeepSWE software engineering agent
Milestones
- 2025
DeepScaleR released
A 1.5B model trained with distributed reinforcement learning.
- 2025
DeepCoder released
A 14B open code reasoning model.
- 2025
rLLM released
An open framework for training language agents with reinforcement learning.
- 2025
DeepSWE released
A 32B software engineering agent trained with reinforcement learning.
Open source
RL for language agents.
Async pipelined verl.
R2E-Gym.
DeepScaleR.
DeepCoder.
- DeepScaleR-Preview-Datasetdataset
Training data.
Notable releases
Openness
Open weights, data and code.
People
- Michael LuoCore contributor
- Sijun TanCore contributor
- Ion StoicaAdvisor
Listed as advisor on the project site.
- Raluca Ada PopaAdvisor
- Li Erran LiAdvisor
Listed from the lab's own pages and public profiles.
Contact and community
Contact details and links are on the lab's website (https://agentica-project.com); code questions go through issues on their GitHub organisation.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



