We-Math: leaderboard
We-Math benchmark evaluating visual mathematical reasoning, reporting strict average accuracy across problem-solving sub-skills.
Metric: Avg Strict (%). Source: we-math.github.io. Saturation forecast: Estimated already saturated. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 78 | #152 |
| 2 | Doubao-1.5-Pro | 65.7 | #287 |
| 3 | GPT-4.1 (2025-04-14) | 55.5 | #250 |
| 4 | GPT-4o | 50.6 | #343 |
| 5 | InternVL3-78B | 46.1 | #352 |
| 6 | InternVL3-8B | 39.5 | #620 |
| 7 | Qwen 2 VL 72B | 36.57 | #460 |
| 8 | Qwen 2.5 VL 7B | 36.2 | #643 |
| 9 | URSA-8B | 32.8 | #811 |
| 10 | InternVL2-8B | 26.57 | #817 |
| 11 | Gemini 1.5 Pro | 26.38 | #445 |
| 12 | Qwen 2 VL 7B | 25.62 | #750 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=we-math · How It Works · Data refreshed daily, snapshot 2026-10-20.