EBR-bench
Re-run by Epoch AINo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Run independently by Epoch AI, with public logs. Our own re-runs are not built yet.
- Score shown
- Epoch's column "Best score (across scorers)", as a percentage, higher is better; random guessing scores about 0.0%. Error bars as the source gives them.
- Results
- 33 in all: 3 for open-weight models, 30 for closed ones. 2 linked to a model page here.
Best open-weight result: mistral-large-4-0-oss_high, 18.1%, number 23 of 33 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | gpt-6.2-astra_maxOpenAI | closed | effort max | 92.4%±4.9 | 27 Oct 2026 | Epoch AI |
| 2 | claude-fable-5-2_maxAnthropic | closed | effort max | 90.5%±1.5 | 6 Nov 2026 | Epoch AI |
| 3 | gemini-4-carbon_highGoogle DeepMind | closed | effort high | 87.6%±6.8 | 5 Nov 2026 | Epoch AI |
| 4 | gpt-6.1-terra_maxOpenAI | closed | effort max | 78.1%±5.1 | 26 Oct 2026 | Epoch AI |
| 5 | gpt-6-astra_maxOpenAI | closed | effort max | 76.2%±8.4 | 30 Aug 2026 | Epoch AI |
| 6 | claude-opus-5-5_maxAnthropic | closed | effort max | 71.4%±7.1 | 22 Sep 2026 | Epoch AI |
| 7 | gemini-4-hydrogen_xhighGoogle DeepMind | closed | effort xhigh | 70.5%±6.3 | 13 Nov 2026 | Epoch AI |
| 8 | gemini-4-argon_highGoogle DeepMind | closed | effort high | 64.8%±4.9 | 13 Oct 2026 | Epoch AI |
| 9 | claude-fable-5-1_maxAnthropic | closed | effort max | 57.1%±6.0 | 8 Sep 2026 | Epoch AI |
| 10 | gpt-6.1-sol_maxOpenAI | closed | effort max | 54.3%±5.8 | 29 Sep 2026 | Epoch AI |
| 11 | gpt-6-sol_maxOpenAI | closed | effort max | 53.3%±4.6 | 22 Sep 2026 | Epoch AI |
| 12 | claude-opus-5_maxAnthropic | closed | effort max | 45.7%±3.6 | 5 Sep 2026 | Epoch AI |
| 13 | gpt-5.6-sol_maxOpenAI | closed | effort max | 44.8%±4.2 | 4 Sep 2026 | Epoch AI |
| 14 | claude-haiku-5-5_maxAnthropic | closed | effort max | 43.8%±1.8 | 9 Oct 2026 | Epoch AI |
| 15 | claude-fable-5_maxAnthropic | closed | effort max | 39.5%±2.0 | 17 Jul 2026 | Epoch AI |
| 16 | gpt-5.5_xhighOpenAI | closed | effort xhigh | 34.3%±2.0 | 27 Jul 2026 | Epoch AI |
| 17 | muse-wonder-1.0_maxMeta AI | closed | effort max | 31.4%±4.2 | 12 Oct 2026 | Epoch AI |
| 18 | grok-4.6_xhighxAI | closed | effort xhigh | 30.5%±1.6 | 18 Aug 2026 | Epoch AI |
| 19 | claude-opus-4-8_maxAnthropic | closed | effort max | 28.6%±1.4 | 7 Aug 2026 | Epoch AI |
| 20 | gpt-5.4-2026-03-05_xhighOpenAI | closed | effort xhigh | 25.4% | 25 Jun 2026 | Epoch AI |
| 21 | gpt-5.2-2025-12-11_xhighOpenAI | closed | effort xhigh | 23.0% | 26 Jun 2026 | Epoch AI |
| 22 | claude-opus-4-7_maxAnthropic | closed | effort max | 19.0% | 30 Jun 2026 | Epoch AI |
| 23 | mistral-large-4-0-oss_highMistral AIno page match | open | effort high | 18.1%±3.2 | 2 Nov 2026 | Epoch AI |
| 24 | claude-opus-4-5-20251101_128KAnthropic | closed | default | 14.3% | 25 Jun 2026 | Epoch AI |
| 25 | gemini-3.1-pro-previewGoogle DeepMind | closed | default | 14.3% | 25 Jun 2026 | Epoch AI |
| 26 | claude-opus-4-6_maxAnthropic | closed | effort max | 12.7% | 29 Jun 2026 | Epoch AI |
| 27 | gpt-5-2025-08-07_highOpenAI | closed | effort high | 12.7% | 29 Jun 2026 | Epoch AI |
| 28 | zai-org/GLM-5.2Epoch: glm-5.2_max, Z.ai (Zhipu AI) | open | effort max | 9.5% | 29 Jun 2026 | Epoch AI |
| 29 | qwen3.7-maxAlibaba | closed | default | 9.5% | 25 Jun 2026 | Epoch AI |
| 30 | claude-opus-4-1-20250805Anthropic | closed | default | 7.9% | 25 Jun 2026 | Epoch AI |
| 31 | gemini-3.5-flash_highGoogle DeepMind | closed | effort high | 4.8% | 25 Jun 2026 | Epoch AI |
| 32 | claude-sonnet-4-5-20250929Anthropic | closed | default | 2.4% | 25 Jun 2026 | Epoch AI |
| 33 | moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshot | open | default | 2.4% | 25 Jun 2026 | Epoch AI |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method