Eleusis Benchmark: leaderboard
Eleusis inductive-reasoning benchmark, scoring how well models infer a hidden rule from sequential examples (average floored score).
Metric: Avg Floored Score. Source: huggingface.co. Saturation forecast: Around April 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 17.26 | #66 (Gemini 3 Pro (Preview)) |
| 2 | Claude Opus 4.5 | 16.97 | #83 |
| 3 | Kimi K2 | 16.18 | #243 |
| 4 | GLM-4.7 | 15.64 | #192 |
| 5 | Grok 4.1 Fast (Reasoning) | 15.15 | #217 (Grok 4.1 Fast) |
| 6 | DeepSeek V3.2 | 15.05 | #205 |
| 7 | GPT-5.2 (High) | 15.05 | #110 (GPT-5.2) |
| 8 | Gemini 3 Flash (Preview) (High) | 14.05 | #82 (Gemini 3 Flash (Preview)) |
| 9 | GPT-5 Mini (Medium) | 13.49 | #186 (GPT-5 Mini) |
| 10 | DeepSeek R1 | 13.28 | #251 |
| 11 | Gemini 3 Flash (Preview) (Low) | 13.13 | #82 (Gemini 3 Flash (Preview)) |
| 12 | GPT-OSS-120B | 12.87 | #341 |
| 13 | GPT-OSS-20B | 11.88 | #518 |
| 14 | Claude Haiku 4.5 | 11.46 | #280 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=eleusis-benchmark · How It Works · Data refreshed daily, snapshot 2026-10-20.