EBR-bench leaderboard
- GPT-6 Astra leads
- 76.2%
- models measured
- 26
- lead over #2
- +4.8 pts
- median
- 27.0%
Leader on EBR-bench · read October 21
On October 21, 2026, 26 of our catalog's models had a EBR-bench result, and OpenAI GPT-6 Astra led with 76.2%, ahead of Anthropic Claude Opus 5.5 at 71.4%.
Epoch AI collects its results; its benchmarking hub describes the tasks and links the authors.
Role in the index
EBR-bench is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 76.2% | 85.1 | $50.00 | |
| 2 | Claude Opus 5.5Anthropic | 71.4% | 88.8 | $20.00 | |
| 3 | Gemini 4 ArgonGoogle | 64.8% | 81.0 | $10.00 | |
| 4 | Claude Fable 5.1Anthropic | 57.1% | 84.2 | $50.00 | |
| 5 | GPT-6.1 SolOpenAI | 54.3% | 82.4 | $10.00 | |
| 6 | GPT-6 SolOpenAI | 53.3% | 76.0 | $10.00 | |
| 7 | Claude Opus 5Anthropic | 45.7% | 83.0 | $25.00 | |
| 8 | GPT-5.6 SolOpenAI | 44.8% | 73.5 | $20.00 | |
| 9 | Claude Haiku 5.5Anthropic | 43.8% | 70.7 | $0.50 | |
| 10 | Claude Fable 5Anthropic | 39.5% | 79.3 | $50.00 | |
| 11 | GPT-5.5OpenAI | 34.3% | 71.3 | $30.00 | |
| 12 | Grok 4.6xAI | 30.5% | 65.7 | $6.00 | |
| 13 | Claude Opus 4.8Anthropic | 28.6% | 65.0 | $25.00 | |
| 14 | GPT-5.4OpenAI | 25.4% | 64.4 | $15.00 | |
| 15 | GPT-5.2OpenAI | 23.0% | 55.9 | $14.00 | |
| 16 | Claude Opus 4.7Anthropic | 19.0% | 67.5 | $25.00 | |
| 17 | Claude Opus 4.5Anthropic | 14.3% | 44.0 | $25.00 | |
| 18 | Gemini 3.1 Pro PreviewGoogle | 14.3% | 32.3 | $12.00 | |
| 19 | Claude Opus 4.6Anthropic | 12.7% | 57.2 | $25.00 | |
| 20 | GPT-5OpenAI | 12.7% | 38.8 | $10.00 | |
| 21 | Qwen3.7 MaxQwen | 9.5% | 54.1 | $4.425 | |
| 22 | GLM 5.2Z.ai | 9.5% | 61.1 | $4.40 | |
| 23 | Claude Opus 4.1Anthropic | 7.9% | Not benchmarked | $75.00 | |
| 24 | Gemini 3.5 FlashGoogle | 4.8% | 39.8 | $9.00 | |
| 25 | Claude Sonnet 4.5Anthropic | 2.4% | 39.5 | $15.00 | |
| 26 | Kimi K2.6Moonshot AI | 2.4% | 52.1 | $4.00 |
Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out.