Verifiable environments and expert trajectories for training agents.
Catalog
| Category | Offering | Benchmark / capability | One unit | Samples |
|---|---|---|---|---|
| Coding and terminal agents | Terminal-Bench tasks | Improves: Terminal-Bench 4.0 | One Harbor task directory with its verifier and reference solution | 3 sample tasks |
| CursorBench-style coding tasks | Improves: CursorBench 4.0 | One task environment plus one agent trajectory | 8 sample tasks | |
| DeepSWE-style long-horizon SWE tasks | Improves: DeepSWE v1.1 | One Harbor task plus agent trajectories | 8 sample tasks | |
| FrontierCode-style mergeable-code tasks | Improves: FrontierCode 1.1 | One task environment plus agent trajectories | 7 sample tasks | |
| Multi-turn coding agent trajectories | Improves: SWE | One session: a JSONL file, one line per API call | 5 session excerpts | |
| Computer-use agents | Long-horizon computer-use workflows | Improves: OSWorld 2.0 | One task environment plus one expert trajectory | 7 sample tasks |
| 3D, graphics and games | Blender 3D scene building | Improves: 3D modeling in Blender | One finished .blend scene and its full request-level trajectory | 10 samples |
| Browser 3D game development trajectories | Improves: Long-horizon web and game development | One game: source, per-round requests and full trajectory | 3 sample games |
Built to your spec
Specify any of the following; anything else is open for discussion.
- Benchmark and format
- Which benchmark to target, and the format the data is delivered in.
- Domains and mix
- Fields, programming languages, apps and task types, and their proportions.
- Difficulty
- The reference model and the pass-rate band to hit.
- Trajectories
- The model that produces them, the agent harness and the tools.
- Volume and schedule
- How much data you need and when it is delivered.
- Rights
- Exclusivity, retention and publication.
Tell us what your model needs to get better at.
We reply with samples and a specification.