Benchmark
SuperCLUE 数理推理
Scores are percentages.
Benchmark sourceScores from SuperCLUE data.
- Models measured
- 20
- Scores recorded
- 20
- Sources
- 1
- IQ weight
- 0.70
Score distribution
- 63.9%
- Mean
- 65.8%
- Median
- 6.7%
- Std dev
- 57.9%
- 10th percentile
- 71.1%
- 90th percentile
- 50%
- Min
This benchmark has a weight of 0.70 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT-6.1 Sol· SuperCLUE data | 76.3% |
| 2 | Qwen4 Max· SuperCLUE data | 73.7% |
| 3 | DeepSeek V4.1 Flash 1017· SuperCLUE data | 71.1% |
| 4 | Gemini 4 Argon· SuperCLUE data | 71.1% |
| 5 | Grok 4.7· SuperCLUE data | 68.4% |
| 6 | Hy 4· SuperCLUE data | 68.4% |
| 7 | Qwen3.8 Max· SuperCLUE data | 68.4% |
| 8 | Spark-X2.6· SuperCLUE data | 68.4% |
| 9 | GLM 5.3· SuperCLUE data | 65.8% |
| 10 | MiniMax M3.1 Flash· SuperCLUE data | 65.8% |
| 11 | Kimi K3.1 Pro· SuperCLUE data | 63.2% |
| 12 | MiniMax M3· SuperCLUE data | 63.2% |
| 13 | Doubao Seed 2.1 Pro· SuperCLUE data | 60.5% |
| 14 | GLM 5.3 Flash· SuperCLUE data | 57.9% |
| 15 | Hy4 preview· SuperCLUE data | 57.9% |
| 16 | MiMo V2.6 Flash· SuperCLUE data | 57.9% |
| 17 | Qwen3.8 Flash· SuperCLUE data | 57.9% |
| 18 | Step 5 Preview· SuperCLUE data | 57.9% |
| 19 | LongCat 2.5 Preview· SuperCLUE data | 55.3% |
| 20 | MiMo V2.6 Pro· SuperCLUE data | 50% |