KernelBench Mega — B200 Best Decode Speedup
Code · 2026-06-18
KernelBench Mega best valid decode speedup on the B200 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 |
|---|---|
| Opus 4.8 | 19.4 |
| GPT-5.5 | 9.4 |
| GLM-5.2 | 7.3 |
| MiniMax M3 | 4.2 |
| DeepSeek-V4-Pro | 1.6 |
| Composer 2.5 | 1.2 |