KernelBench Hub - Mega: leaderboard
kernelbench.com Mega suite (independent of Stanford KernelBench) — agentic GPU megakernel building; best speedup over a reference megakernel across RTX PRO 6000, H100, and B200 runs.
Metric: Best Speedup vs Reference (x). Source: kernelbench.com. Saturation forecast: Around 2029. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 35.46 | #3 |
| 2 | Claude Sonnet 5.5 | 32.13 | #11 |
| 3 | Gemini 4 Argon (High) | 27.03 | #8 (Gemini 4 Argon) |
| 4 | Claude Fable 5 | 24.61 | #16 |
| 5 | Claude Fable 5.1 | 22.95 | #9 |
| 6 | GLM-5.3 | 19.43 | #39 |
| 7 | Kimi K3 | 18.09 | #30 |
| 8 | DeepSeek V4.1 Flash | 17.1 | #36 |
| 9 | GLM-5.3 Flash | 13.64 | #65 |
| 10 | Grok 4.7 | 6.38 | #25 |
| 11 | GPT-6.1 Sol | 4.99 | #7 |
| 12 | Gemini 3.8 Flash (High) | 2.74 | #21 (Gemini 3.8 Flash) |
| 13 | GPT-5.6 Sol | 2.64 | #19 |
| 14 | GPT-6 Luna | 2.48 | #56 |
| 15 | Muse Spark 1.3 | 2.18 | #24 |
No result here: #6 GPT-6 Astra, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=kernelbench-hub-mega · How It Works · Data refreshed daily, snapshot 2026-10-20.