Kernel Bench L3: leaderboard
Kernel optimization benchmark reporting median speedup over a PyTorch eager reference and the fraction of problems faster than torch.compile.
Metric: Median speedup (x). Source: qwen.ai. Source: benchmarklist.com. Saturation forecast: Around October 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Max) | 2.63 | #63 (Claude Opus 4.6) |
| 2 | Qwen 3.7 Plus | 2.06 | #92 |
| 3 | GLM-5.1 (Thinking) | 2 | #97 (GLM-5.1) |
| 4 | Kimi K2.6 (Thinking) | 1.41 | #104 (Kimi K2.6) |
| 5 | DeepSeek V4 Pro (Max) | 1.07 | #102 (DeepSeek V4 Pro) |
| 6 | Qwen 3.6 Plus | 1.03 | #120 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kernel-bench-l3 · How It Works · Data refreshed daily, snapshot 2026-10-20.