EBR-bench v3
Games · 2026-08-11
Earthborne Rangers objective completion using the best of the final two playthroughs and the better result across two context-compaction settings. This August 11 protocol retains the original card set and differs from the later card-ban runs.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 39.5 |
| GPT-5.5 | 34.3 |
| Grok 4.6 | 30.5 |
| Opus 4.8 | 28.6 |