Vals AI MATH 500: leaderboard
MATH-500 competition-math subset run independently by Vals AI on frontier models; accuracy (%).
Metric: Accuracy (%). Source: www.vals.ai. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 60 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 96.4 | #66 (Gemini 3 Pro (Preview)) |
| 2 | Grok 4 | 96.2 | #181 |
| 3 | GPT-5 | 96 | #98 |
| 4 | GPT-5 (2025-08-07) (High) | 96 | |
| 5 | Claude Opus 4.1 (20250805) (Thinking) | 95.4 | #132 (Claude Opus 4.1 (20250805)) |
| 6 | Gemini 2.5 Pro (Preview 03-25) | 95.2 | #123 |
| 7 | GPT-OSS-120B | 94.8 | #341 |
| 8 | GPT-5 Mini (2025-08-07) (High) | 94.8 | #172 (GPT-5 Mini (2025-08-07)) |
| 9 | Qwen 3 235B A22B | 94.6 | #313 |
| 10 | O3 (2025-04-16) (High) | 94.6 | #121 (O3 (2025-04-16)) |
| 11 | GPT-OSS-20B | 94.2 | #518 |
| 12 | Kimi K2 | 94.2 | #243 |
| 13 | O4 Mini (2025-04-16) (High) | 94.2 | #179 (O4 Mini (2025-04-16)) |
| 14 | Grok 3 Mini (High) | 94.2 | #281 (Grok 3 Mini) |
| 15 | GLM-4.5 | 94 | #275 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-math-500 · How It Works · Data refreshed daily, snapshot 2026-10-20.