Reasoning
EBR-bench Leaderboard
EBR-bench, from Epoch AI, tests whether models learn from experience by having them play the board game Earthborne Rangers over repeated playthroughs, with the rulebook, card database and a persistent note-taking tool. Scores stay roughly flat across plays and well below an expert human baseline.
Source: epoch2 open models ranked+25 proprietaryData through Oct 2026
All models ranked on EBR-bench
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | GPT 6 Astra Max · proprietary | 76.2% | API only |
| 2 | Claude Opus 5.5 Max · proprietary | 71.4% | API only |
| 3 | Gemini 4 Argon (high) · proprietary | 64.8% | API only |
| 4 | Claude Fable 5.1 Max · proprietary | 57.1% | API only |
| 5 | GPT 6.1 Sol Max · proprietary | 54.3% | API only |
| 6 | GPT 6 Sol Max · proprietary | 53.3% | API only |
| 7 | Claude Opus 5 Max · proprietary | 45.7% | API only |
| 8 | GPT 5.6 Sol Max · proprietary | 44.8% | API only |
| 9 | Claude Haiku 5.5 Max · proprietary | 43.8% | API only |
| 10 | Claude Fable 5 Max · proprietary | 39.5% | API only |
| 11 | GPT 5.5 (xhigh) · proprietary | 34.3% | API only |
| 12 | Muse Wonder 1.0 Max · proprietary | 31.4% | API only |
| 13 | Grok 4.6 (xhigh) · proprietary | 30.5% | API only |
| 14 | Claude Opus 4.8 Max · proprietary | 28.6% | API only |
| 15 | GPT 5.4 (Mar 05, 2026, xhigh) · proprietary | 25.4% | API only |
| 16 | GPT 5.2 (Dec 11, 2025, xhigh) · proprietary | 23.0% | API only |
| 17 | Claude Opus 4.7 Max · proprietary | 19.1% | API only |
| 18 | Claude Opus 4.5 (Nov 01, 2025, 128K) · proprietary | 14.3% | API only |
| 19 | Gemini 3.1 Pro Preview · proprietary | 14.3% | API only |
| 20 | Claude Opus 4.6 Max · proprietary | 12.7% | API only |
| 21 | GPT 5 (Aug 07, 2025, high) · proprietary | 12.7% | API only |
| 22 | GLM 5.2 · 753.3B | 9.5% | ~456 GB |
| 23 | Qwen3.7 Max · proprietary | 9.5% | API only |
| 24 | Claude Opus 4.1 (Aug 05, 2025) · proprietary | 7.9% | API only |
| 25 | Gemini 3.5 Flash (high) · proprietary | 4.8% | API only |
| 26 | Claude Sonnet 4.5 (Sep 29, 2025) · proprietary | 2.4% | API only |
| 27 | Kimi K2.6 · 1026.9B | 2.4% | ~620 GB |
EBR-bench: frequently asked questions
- What is the best open LLM on EBR-bench?
- GLM 5.2 is the top open model on EBR-bench, scoring 9.5%. Among all models tested — including proprietary ones — it ranks #22. The top model overall is GPT 6 Astra Max (OpenAI) at 76.2%.
- Can open models match proprietary models on EBR-bench?
- Not quite on EBR-bench: the strongest proprietary model (GPT 6 Astra Max) scores 76.2%, ahead of the best open model (GLM 5.2) at 9.5% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.