AIME 2026
AIME 2026 is run by MathArena (ETH Zurich SRI Lab). It has ranked 32 models, of which the catalogue holds 25, scoring from 91.67 to 100 on Accuracy (± 95% CI).
measured by MathArena (ETH Zurich SRI Lab) · the board itself
| Place | Model | Metric | Score |
|---|---|---|---|
| 1stof 32 | GPT-5.5 | Accuracy (± 95% CI) (xhigh) | 100 |
| 1stof 32 | Claude Opus 4.8 | Accuracy (± 95% CI) (max) | 100 |
| 3rdof 32 | GPT-5.4 | Accuracy (± 95% CI) (xhigh) | 99.17 |
| 4thof 32 | Gemini 3.1 Pro Preview | Accuracy (± 95% CI) | 98.33 |
| 4thof 32 | GPT-5.2 | Accuracy (± 95% CI) (high) | 98.33 |
| 6thof 32 | Step 3.5 Flash | Accuracy (± 95% CI) | 96.67 |
| 6thof 32 | Gemini 3 Flash Preview | Accuracy (± 95% CI) | 96.67 |
| 6thof 32 | GLM-5 | Accuracy (± 95% CI) | 96.67 |
| 6thof 32 | Gemini 3.6 Flash | Accuracy (± 95% CI) | 96.67 |
| 6thof 32 | DeepSeek V4 Pro | Accuracy (± 95% CI) (max) | 96.67 |
| 6thof 32 | Kimi K3 | Accuracy (± 95% CI) (think) | 96.67 |
| 6thof 32 | Claude Opus 4.6 | Accuracy (± 95% CI) (high) | 96.67 |
| 13thof 32 | GLM 5.1 | Accuracy (± 95% CI) | 95.83 |
| 13thof 32 | Kimi K2.6 | Accuracy (± 95% CI) (think) | 95.83 |
| 13thof 32 | DeepSeek V4 Flash | Accuracy (± 95% CI) (max) | 95.83 |
| 13thof 32 | Kimi K2.5 | Accuracy (± 95% CI) (think) | 95.83 |
| 13thof 32 | Claude Opus 4.7 | Accuracy (± 95% CI) (xhigh) | 95.83 |
| 18thof 32 | Step 3.7 Flash | Accuracy (± 95% CI) | 95 |
| 18thof 32 | Gemini 3.5 Flash | Accuracy (± 95% CI) | 95 |
| 20thof 32 | Grok 4.1 Fast Reasoning | Accuracy (± 95% CI) | 94.17 |
| 20thof 32 | Qwen3.5 397B A17B | Accuracy (± 95% CI) | 94.17 |
| 20thof 32 | DeepSeek v3.2 | Accuracy (± 95% CI) (think) | 94.17 |
| 23rdof 32 | Qwen3.5-35B-A3B | Accuracy (± 95% CI) | 93.33 |
| 24thof 32 | Qwen3.5 9B | Accuracy (± 95% CI) | 92.5 |
| 25thof 32 | Nemotron 3 Super | Accuracy (± 95% CI) | 91.67 |
Read from the board on 2026-08-25