Epoch AI - ECI: leaderboard
Epoch Capabilities Index, a composite that fits an item response theory model to results from over 50 benchmarks, estimating benchmark difficulty and model capability together. Results come from Epoch's own runs, independent leaderboards and developer reports. The scale is unbounded and anchored so Claude 3.5 Sonnet is 130 and GPT-5 is 150, higher is better.
Metric: ECI Score. Source: epoch.ai. Saturation forecast: Around September 2027. 277 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 166.82 | #3 |
| 2 | GPT-6 Astra | 166.41 | #6 |
| 3 | GPT-6.1 Sol | 166.32 | #7 |
| 4 | Claude Fable 5.1 | 164.65 | #9 |
| 5 | Gemini 4 Argon | 164.56 | #8 |
| 6 | Claude Sonnet 5.5 | 164.45 | #11 |
| 7 | Claude Opus 5 | 162.6 | #12 |
| 8 | GPT-6 Sol | 162.55 | #14 |
| 9 | Claude Fable 5 | 162.13 | #16 |
| 10 | GPT-5.5 Pro | 161.97 | #17 |
| 11 | GPT-5.6 Sol | 161.73 | #19 |
| 12 | GPT-5.6 Terra | 159.43 | #41 |
| 13 | GPT-5.4 Pro (xHigh) | 159 | #32 (GPT-5.4 Pro) |
| 14 | Claude Haiku 5.5 | 158.98 | #40 |
| 15 | GPT-5.5 | 158.95 | #31 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-eci · How It Works · Data refreshed daily, snapshot 2026-10-20.