MathArena - APEX Shortlist 2025: leaderboard
Curated shortlist of the hardest competition problems from 2025 final-answer olympiads, probing the ceiling of AI mathematical reasoning.
Metric: Accuracy (%). Source: matharena.ai. Saturation forecast: Around January 2027. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 98.4 | #31 (GPT-5.5) |
| 2 | Gemini 3.1 Pro (Preview) | 92.02 | #58 |
| 3 | Kimi K3 (Thinking) | 92.02 | #30 (Kimi K3) |
| 4 | Claude Opus 4.8 (Max) | 90.43 | #37 (Claude Opus 4.8) |
| 5 | DeepSeek V4 Flash (Max) | 89.36 | #137 (DeepSeek V4 Flash) |
| 6 | DeepSeek V4 Pro (Max) | 87.77 | #102 (DeepSeek V4 Pro) |
| 7 | Claude Opus 4.6 (High) | 86.7 | #63 (Claude Opus 4.6) |
| 8 | Gemini 3.5 Flash | 82.45 | #60 |
| 9 | GPT-5.4 (xHigh) | 81.38 | #79 (GPT-5.4) |
| 10 | GPT-5.2 (High) | 79.26 | #110 (GPT-5.2) |
| 11 | Kimi K2.6 (Thinking) | 77.13 | #104 (Kimi K2.6) |
| 12 | Step 3.7 Flash | 76.6 | #236 |
| 13 | GLM-5.1 | 71.81 | #97 |
| 14 | Gemini 3.6 Flash | 71.28 | #51 |
| 15 | Step 3.5 Flash | 70.74 | #270 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=matharena-apex-shortlist-2025 · How It Works · Data refreshed daily, snapshot 2026-10-20.