RL environments for frontier models
Long-horizon tasks from real expert work, run in real tools and graded automatically, ready for your training stack.
How our environments work
Containerized and reproducible, with adapters for the agent harnesses you already run.
- 01
A real task
Long-horizon work drawn from how experts actually operate.
- 02
Real tools
The agent searches, reads, runs and edits, just as a person would.
- 03
An automatic verdict
A verifier grades the answer and its evidence, outside the agent’s reach.
Built with domain experts
Every task is grounded in real work and checked by specialists in the field.
- 01Domain expertsPhDs, engineers, lawyers, clinicians and analysts.
- 02Generated at scaleFresh tasks with known answers, reviewed by experts.
- 03Licensed real dataReal-world sources through data partnerships.
How we build every environment
The same bar in every domain, from the first task to the last.
- 01
Verifiable
Automatic verifiers check every answer and the evidence behind it.
- 02
Frontier-hard
Calibrated so today’s best models are challenged, and can still learn.
- 03
Uncontaminated
Test sets are held out and never used for training.
- 04
Expert-built
Designed and reviewed by people who do the work.
Domains we cover
Built with the people who do the work in each field.
- Software engineeringRepositories, tests and tooling
- MathematicsProofs and quantitative reasoning
- Life sciencesBiology and chemistry
- Physical sciencesPhysics and engineering
- FinanceReports, models and filings
- LawContracts and case work
- MedicineRecords and guidelines
- Computer useBrowsers, apps and documents