Map where models do well and where they slip.
teleon.dev/docs/frontier keeps a capability matrix across tasks and harnesses, runs public and custom benchmarks, and scans for cliffs: similar tasks where results drop sharply.
Merged-cell tables fall sharply next to simple tables, which marks a cliff worth fixing.
What it does
Capability matrix
Pass rates by task family and harness, kept current.
Public and custom
Public benchmarks sit beside your own holdouts.
Cliffs become fixes
A sharp drop between neighbours seeds a repair task.
The contract
A frontier cell names the task family, the harness, pass@k and calibration.
/v1/frontier/jagged-mapteleon_jagged_frontier_cartographer/v1pass_at_k_vs_pass_hat_k
Endpoints
| Method | Path | Access | What it does | Status |
|---|---|---|---|---|
GET | /v1/frontier/jagged-map | RING-3 | Read cliffs for a task family | Listed |
GET | /v1/arena/compare | RING-3 | Compare harnesses on a suite | Listed |
Listed paths appear in the service’s published files today. In the one-host plan every path here is served under api.teleon.dev/v1/frontier.
How it connects
For agents
Give your agent the read-only guide first. It reports back before any write.
Words on this page
- Jagged frontier
- The uneven edge of what models do well: strong on some tasks, weak on close neighbours.
- pass@k
- The share of tasks solved at least once in k tries.
- Holdout
- Test data kept apart from training and tuning, so scores on it show real performance.