KernelBench Mega — RTX PRO 6000 Best Decode Speedup
Code · 2026-06-18
KernelBench Mega best valid decode speedup on the RTX PRO 6000 Blackwell board. The live headline metric uses the Kimi-Linear W4A16 decode problem and excludes the sparsely attempted RL Grid PPO cell from the performance aggregate.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 18.7 |
| Kimi K3 | 18.1 |
| Opus 4.8 | 14.4 |
| GLM-5.2 | 11.1 |
| GPT-5.5 | 4.3 |
| Sonnet 5 | 4.0 |
| GPT-5.6 Sol | 2.6 |
| MiniMax M3 | 2.6 |
| Kimi K2.7 Code | 2.6 |
| Composer 2.5 | 2.5 |
| DeepSeek-V4-Pro | 2.1 |
| Grok 4.5 | 0.8 |