Benchmarks · math
Apex Shortlist 2025 (MathArena)
47 final-answer problems from 2025 competitions on which at least one of two reference models (Grok 4 Fast, GPT-5 mini) got at least one of four attempts wrong, run by MathArena. Maintained by MathArena. Score: Accuracy (average performance on the competition) with a 95% confidence interval (normal approximation).
33 models · 33 results · data as of 2026-10-29 · matharena.ai ↗
Leaderboard
GPT-5.5 xhigh
All results
33 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5.5 | xhigh | 98.4% |
| 2 | Gemini 3.1 Pro | high | 92.0% |
| 3 | Kimi K3 | max | 92.0% |
| 4 | Opus 4.8 | max | 90.4% |
| 5 | DeepSeek V4 Flash | max | 89.4% |
| 6 | DeepSeek V4 Pro | max | 87.8% |
| 7 | Opus 4.6 | high | 86.7% |
| 8 | Gemini 3.5 Flash | xhigh | 82.5% |
| 9 | GPT-5.4 | xhigh | 81.4% |
| 10 | GPT-5.2 | high | 79.3% |
| 11 | Kimi K2.6 | max | 77.1% |
| 12 | GLM 5.1 | max | 71.8% |
| 13 | Gemini 3.6 Flash | xhigh | 71.3% |
| 14 | DeepSeek V3.2 | high | 69.7% |
| 15 | Gemini 3 Flash | high | 68.6% |
| 16 | GLM 5 | high | 68.6% |
| 17 | GLM 5.2 | max | 68.1% |
| 18 | Gemini 3 Pro | high | 66.5% |
| 19 | Opus 4.7 | xhigh | 63.8% |
| 20 | Qwen 3.5 397B-A17B | max | 60.1% |
| 21 | Grok 4.1 Fast | high | 57.9% |
| 22 | Grok 4 | high | 57.5% |
| 23 | Kimi K2.5 | max | 57.5% |
| 24 | GPT-5.1 | high | 55.9% |
| 25 | Grok 4 Fast | none | 55.3% |
| 26 | Qwen 3.5 27B | high | 52.1% |
| 27 | Kimi K2 | medium | 47.3% |
| 28 | gpt-oss-120b | high | 45.2% |
| 29 | Qwen 3.5 35B-A3B | low | 44.7% |
| 30 | GPT-5 Mini | high | 39.9% |