MathArena - AIME 2026: leaderboard
American Invitational Mathematics Examination 2026, providing contamination-free evaluation of competition math reasoning.
Metric: Accuracy (%). Source: matharena.ai. Saturation forecast: Estimated already saturated. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 100 | #31 (GPT-5.5) |
| 2 | Claude Opus 4.8 (Max) | 100 | #37 (Claude Opus 4.8) |
| 3 | Best Expert | 100 | |
| 4 | GPT-5.4 (xHigh) | 99.17 | #79 (GPT-5.4) |
| 5 | Gemini 3.1 Pro (Preview) | 98.33 | #58 |
| 6 | GPT-5.2 (High) | 98.33 | #110 (GPT-5.2) |
| 7 | Gemini 3 Flash | 96.67 | #99 |
| 8 | GLM-5 | 96.67 | #141 |
| 9 | Gemini 3.6 Flash | 96.67 | #51 |
| 10 | Step 3.5 Flash | 96.67 | #270 |
| 11 | Claude Opus 4.6 (High) | 96.67 | #63 (Claude Opus 4.6) |
| 12 | DeepSeek V4 Pro (Max) | 96.67 | #102 (DeepSeek V4 Pro) |
| 13 | Kimi K3 (Thinking) | 96.67 | #30 (Kimi K3) |
| 14 | GLM-5.1 | 95.83 | #97 |
| 15 | Kimi K2.5 (Thinking) | 95.83 | #144 (Kimi K2.5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=matharena-aime-2026 · How It Works · Data refreshed daily, snapshot 2026-10-20.