Benchmarks
347 sources · 1,258 models × 3,038 benchmarks · 158,912 scoresEBR-bench tests whether AI agents improve at the unfamiliar, complex card-based campaign game Earthborne Rangers through repeated play and persistent note-taking. Results must use a versioned child benchmark because v1 and v2 use incompatible topline aggregation and sampling protocols.
No signal or difficulty estimate. No usable scores.
This is a benchmark-family page. Scores are recorded on the component benchmarks below rather than directly on the parent, because the components use different score scales and directions.
| Metric | Models | Leader | Direction | Date |
|---|---|---|---|---|
| EBR-bench v1 | 13 | GPT-5.425.4 | higher is better | — |
| EBR-bench v4 (card ban, single agent) | 13 | Claude Fable 5.290.5 | higher is better | — |
| EBR-bench v2 | 5 | Claude Opus 545.7 | higher is better | 2026-08-02 |
| EBR-bench v3 | 4 | Claude Fable 539.5 | higher is better | — |