IMO-Bench: leaderboard
IMO-Bench (Google DeepMind) — International Mathematical Olympiad-level suite covering short-answer problem solving and rubric-graded proof writing.
Metric: Advanced ProofBench Accuracy (%). Source: imobench.github.io. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 Pro (xHigh) | 88.1 | #17 (GPT-5.5 Pro) |
| 2 | Gemini 3 Deep Think | 76.7 | #27 |
| 3 | GPT-5.5 (xHigh) | 71.9 | #31 (GPT-5.5) |
| 4 | Gemini 3.1 Pro (Preview) | 49 | #58 |
| 5 | GPT-5.2 (High) | 35.7 | #110 (GPT-5.2) |
| 6 | Gemini 3 Pro | 30 | #80 |
| 7 | GPT-5 Pro | 28.6 | #81 |
| 8 | Claude Opus 4.5 | 23.8 | #83 |
| 9 | Grok 4.1 Fast (Reasoning) | 18.6 | #217 (Grok 4.1 Fast) |
| 10 | GPT-5.1 | 7.1 | #135 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=imo-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.