Benchmarks · math
Apex 2025 (MathArena)
Twelve final-answer problems from 2025 competitions that MathArena selected because strong models of the time failed them; the selection is biased against the four models used to choose it (Grok 4, GPT-5, Gemini 2.5 Pro, GLM 4.5). Maintained by MathArena. Score: Accuracy (average performance on the competition) with a 95% confidence interval (normal approximation).
41 models · 41 results · data as of 2026-10-29 · matharena.ai ↗
Leaderboard
Opus 4.8 max
All results
41 rows| # | Model | Effort | Apex 2025 (MathArena) |
|---|---|---|---|
| 1 | Opus 4.8 | max | 81.3% |
| 2 | GPT-5.5 | xhigh | 80.2% |
| 3 | GPT-5.4 Pro | xhigh | 69.8% |
| 4 | Kimi K3 | max | 65.6% |
| 5 | Gemini 3.1 Pro | high | 60.9% |
| 6 | GPT-5.4 | xhigh | 54.2% |
| 7 | Opus 4.7 | xhigh | 40.6% |
| 8 | Opus 4.6 | high | 34.5% |
| 9 | Gemini 3.5 Flash | xhigh | 32.3% |
| 10 | DeepSeek V4 Pro | max | 28.1% |
| 11 | DeepSeek V4 Flash | max | 27.1% |
| 12 | Gemini 3.6 Flash | xhigh | 26.0% |
| 13 | Kimi K2.6 | max | 24.0% |
| 14 | Gemini 3 Pro | high | 23.4% |
| 15 | GLM 5.2 | max | 19.8% |
| 16 | Gemini 3 Flash | high | 15.6% |
| 17 | GPT-5.2 | high | 13.5% |
| 18 | GLM 5.1 | max | 11.5% |
| 19 | GLM 5 | high | 10.9% |
| 20 | DeepSeek V3.2 | high | 9.4% |
| 21 | Kimi K2.5 | max | 8.8% |
| 22 | Grok 4 Fast | none | 5.2% |
| 23 | Grok 4.1 Fast | high | 5.2% |
| 24 | Qwen 3 235B | 5.2% | |
| 25 | Qwen 3.5 35B-A3B | low | 4.2% |
| 26 | GPT-5 | high | 2.1% |
| 27 | Grok 4 | high | 2.1% |
| 28 | Qwen 3 4B | none | 2.1% |
| 29 | Qwen 3.5 27B | high | 2.1% |
| 30 | Sonnet 4.5 | high | 1.6% |