SuperCLUE General (March 2026) - Math Reasoning: leaderboard
Metric: Score. Source: www.superclueai.com. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 92.44 |
| 2 | DeepSeek V4 Flash (Max) | 89.08 |
| 3 | GPT-5.4 (xHigh) | 88.89 |
| 4 | DeepSeek V4 Pro (Max) | 87.39 |
| 5 | Grok 4.20 Beta (0309) (Reasoning) | 85.71 |
| 6 | Gemini 3 Flash (Preview) (High) | 85.71 |
| 7 | Claude Opus 4.6 (Max) | 85.71 |
| 8 | Qwen 3.5 397B A17B (Thinking) | 84.87 |
| 9 | MiMo-V2-Pro | 84.03 |
| 10 | Qwen 3.5 122B A10B (Thinking) | 82.35 |
| 11 | Kimi K2.5 (Thinking) | 81.51 |
| 12 | Step 3.5 Flash | 80.67 |
| 13 | GPT-OSS-120B (High) | 79.83 |
| 14 | MiniMax-M2.7 | 78.15 |
| 15 | DeepSeek V3.2 (Thinking) | 78.15 |
Interactive version: theaggregate.ai/benchmark?slug=superclue-general-march-2026-math-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-19.