Benchmark
KernelBench Peak Performance
Coding benchmarks compare models on programming tasks. Scores are percentages.
All coding benchmarksScores from KernelBench (kernelbench.com) data.
- Models measured
- 15
- Scores recorded
- 15
- Sources
- 1
- IQ weight
- 1.01
Score distribution
- 25.1%
- Mean
- 24.7%
- Median
- 4.9%
- Std dev
- 18.8%
- 10th percentile
- 33.1%
- 90th percentile
- 16.4%
- Min
This benchmark has a weight of 1.01 for chat models, measured over 15 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 | GLM 5.3hard· KernelBench (kernelbench.com) data | 36.2% |
| 2 | Claude Opus 4.8hard· KernelBench (kernelbench.com) data | 33.1% |
| 3 | GLM 5.2hard· KernelBench (kernelbench.com) data | 29.4% |
| 4 | DeepSeek V4 Flash 0731hard· KernelBench (kernelbench.com) data | 27% |
| 5 | Grok 4.6hard· KernelBench (kernelbench.com) data | 26.2% |
| 6 | Grok 4.5hard· KernelBench (kernelbench.com) data | 26% |
| 7 | MiniMax M3hard· KernelBench (kernelbench.com) data | 25.7% |
| 8 | Claude Fable 5hard· KernelBench (kernelbench.com) data | 24.7% |
| 9 | Qwen3.8 Maxhard· KernelBench (kernelbench.com) data | 24.6% |
| 10 | DeepSeek V4 Prohard· KernelBench (kernelbench.com) data | 23.4% |
| 11 | Claude Opus 5hard· KernelBench (kernelbench.com) data | 23.3% |
| 12 | DeepSeek V4.1 Flash 1017hard· KernelBench (kernelbench.com) data | 21.2% |
| 13 | GPT-5.6 Solhard· KernelBench (kernelbench.com) data | 20.9% |
| 14 | DeepSeek Flashhard· KernelBench (kernelbench.com) data | 18.8% |
| 15 | LongCat 2.0hard· KernelBench (kernelbench.com) data | 16.4% |