Kernel Bench L3

Kernel optimization benchmark reporting median speedup over a PyTorch eager reference and the fraction of problems faster than torch.compile.

6rows
median_speedupprimary metric
2026-05-28sampled

Metadata

ID
kernel_bench_l3
Category
Coding
Release
Unknown

Metrics

Median speedup, Win rate, Median speedup

Latest Results

Provider-published Qwen3.7-Max comparison scores. Rows are marked self-reported and should be interpreted as source claims unless independently reproduced.

Rank Subject Median speedup Model Match Provenance Sampled
1 Claude Opus 4.6 Max 2.63/98% Claude Opus 4.6
anthropic-claude-opus-4.6
 		Self-reported 2026-05-28
2 GLM-5.1 Thinking 2.00/78% GLM GLM 5.1
z-ai-glm-5.1
 		Self-reported 2026-05-28
3 Qwen3.7 Max 1.98/96% Qwen3.7 Max
qwen-qwen3.7-max
 		Self-reported 2026-05-28
4 Kimi K2.6 Thinking 1.41/80% KIMI MoonshotAI: Kimi K2.6
moonshotai-kimi-k2.6
 		Self-reported 2026-05-28
5 DeepSeek V4 Pro Max 1.07/54% DeepSeek V4 Pro
deepseek-deepseek-v4-pro
 		Self-reported 2026-05-28
6 Qwen3.6 Plus 1.03/48% Qwen3.6 Plus
qwen-qwen3.6-plus
 		Self-reported 2026-05-28