Epoch AI - Ebr Bench: leaderboard
A model plays Earthborne Rangers, an obscure campaign board game, through a text harness with a compressed rulebook, card lookup and map tools. A run is eight learning playthroughs and two scored ones, sampled at least ten times per model. The score is the better scored playthrough as a share of the 21 objectives in the first five in-game days, a level reached after practice, not a learning gain; higher is better.
Metric: Score. Source: epoch.ai. Saturation forecast: Around January 2028. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 76.19 | #6 (GPT-6 Astra) |
| 2 | Claude Opus 5.5 (Max) | 71.43 | #3 (Claude Opus 5.5) |
| 3 | Gemini 4 Argon (High) | 64.76 | #8 (Gemini 4 Argon) |
| 4 | Claude Fable 5.1 (Max) | 57.14 | #9 (Claude Fable 5.1) |
| 5 | GPT-6.1 Sol (Max) | 54.29 | #7 (GPT-6.1 Sol) |
| 6 | GPT-6 Sol (Max) | 53.33 | #14 (GPT-6 Sol) |
| 7 | Claude Opus 5 (Max) | 45.71 | #12 (Claude Opus 5) |
| 8 | GPT-5.6 Sol (Max) | 44.76 | #19 (GPT-5.6 Sol) |
| 9 | Claude Haiku 5.5 (Max) | 43.81 | #40 (Claude Haiku 5.5) |
| 10 | Claude Fable 5 (Max) | 39.52 | #16 (Claude Fable 5) |
| 11 | GPT-5.5 (xHigh) | 34.29 | #31 (GPT-5.5) |
| 12 | Muse Wonder (Max) | 31.43 | #15 (Muse Wonder) |
| 13 | Grok 4.6 (xHigh) | 30.48 | #26 (Grok 4.6) |
| 14 | Claude Opus 4.8 (Max) | 28.57 | #37 (Claude Opus 4.8) |
| 15 | GPT-5.4 (2026-03-05) (xHigh) | 25.4 | #59 (GPT-5.4 (2026-03-05)) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-ebr-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.