IMO-AnswerBench
Math · 2025-11-03
IMO-AnswerBench is the short-answer track of Google DeepMind's IMO-Bench: 400 olympiad problems in algebra, combinatorics, geometry and number theory, robustified from past competitions, each with a verifiable final answer graded automatically. Scores report answer accuracy, usually averaged over several sampled runs. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| DeepSeek-V4-Pro | 93.0 |
| Nemotron 3 Ultra 550B A55B | 92.3 |
| GPT-5.4 | 91.4 |
| Kimi K2.6 | 91.1 |
| DeepSeek-V4-Flash | 91.1 |
| Qwen3.7-Max (source unspecified) | 90.0 |
| GLM-5.1 | 86.8 |
| GPT-5.2 | 86.3 |
| Qwen3.7-Plus (source unspecified) | 86.0 |
| DeepSeek-V3.2-Speciale | 84.5 |
| Opus 4.5 | 84.0 |
| Qwen3 Max Thinking (2026-01-23) | 83.9 |
| Qwen3.6 Plus (2026-04-02) | 83.8 |
| Gemini 3 Pro Preview | 83.3 |
| GLM-5 | 82.5 |
| Kimi K2.5 | 81.8 |
| Qwen3.6 27B | 80.8 |
| Qwen3.5-27B | 79.9 |
| Nemotron Cascade 2 30B A3B | 79.3 |
| Qwen3.6 35B A3B | 78.9 |
| Kimi K2 Thinking | 78.6 |
| DeepSeek-V3.2 | 78.3 |
| Nemotron 3 Super 120B A12B | 77.2 |
| GPT-5 | 76.0 |
| DeepSeek-V3.2-Exp | 76.0 |
| Opus 4.6 | 75.3 |
| Gemma 4 31B IT | 74.5 |
| Gemma 4 26B A4B IT | 74.3 |
| Nemotron 3 Nano 30B A3B | 70.4 |
| MiniMax M2.7 | 68.3 |
| Qwen3.5 Plus (2026-02-15) | 68.3 |
| Sonnet 4.5 | 65.9 |
| Qwen3.5-Omni-Plus | 65.5 |
| Qwen3.5-Omni-Flash | 51.5 |
| Kimi K2 Instruct 0905 | 45.8 |