Code
Repair failing tests, implement features and pass hidden-test verifiers in real repositories.
Lab · Environment Hub
2,500+ RL environments. Plus the ones only you have.
The largest open-source RL environment library — code, science, reasoning, tool use — filtered by what your model actually needs.
Science & math
Symbolic and numeric problems with verifiable answers.
Reasoning
Multi-step tasks graded by rubrics and exact-match checks.
Tool use
API calls, browser operations and SQL — graded by state and result verifiers.
[ Private environments ]
Your production failures, turned into training environments.
- 1.1
Capture
Failed traces, corrections and retries are collected from production automatically.
- 1.2
Synthesize
Failures become private RL tasks with auto-generated rubrics — e.g. 1,284 traces → 312 tasks.
- 1.3
Train
Private environments are composed with open ones and trained in isolated sandboxes.
[ Ready? ]
Your model should get smarter while you sleep.
Engineers, at full stride. Point PI at your checkpoint — first improvement in about 48 hours.