KernelBench Mega: leaderboard
KernelBench Mega is a text-only benchmark for coding agents working on full fused GPU megakernels. The cited public board specifically measures Kimi-Linear W4A16 decode speedup over an optimized PyTorch baseline. The pinned metric, Speedup (x), measures runtime acceleration versus that baseline. Source: BenchmarkList’s attributed result table.
Metric: Speedup (x). Source: benchmarklist.com. Saturation forecast: Around November 2027. 53 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 35.46 | #3 |
| 2 | Claude Fable 5 | 35.44 | #20 |
| 3 | Claude Sonnet 5.5 | 32.13 | #14 |
| 4 | Claude Opus 5 | 24.29 | #16 |
| 5 | Claude Fable 5.1 | 22.95 | #10 |
| 6 | GLM-5.3 | 19.43 | #44 |
| 7 | Claude Opus 4.8 | 19.35 | #39 |
| 8 | Kimi K3 | 18.09 | #35 |
| 9 | DeepSeek V4.1 Flash | 17.1 | #37 |
| 10 | GLM-5.3 Flash | 13.64 | #77 |
| 11 | GLM-5.2 | 11.14 | #80 |
| 12 | GPT-5.5 | 9.37 | #33 |
| 13 | Grok 4.7 | 6.38 | #26 |
| 14 | DeepSeek V4 Flash (0731) | 5.24 | #84 |
| 15 | DeepSeek V4 Pro Preview | 4.64 | #150 |
No result here: #6 GPT-6 Astra, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=kernelbench-mega · How It Works · Data refreshed daily, snapshot 2026-11-05.