Multi-Head Self-Attention on CUDA-LLM kernels task suite 1.0 (test) benchmark leaderboard · SOTA2 Research

Multi-Head Self-Attention on CUDA-LLM kernels task suite 1.0 (test)

6.85Execution TimeOptiML

Evaluation Results

Method	Links
OptiML2026.02		6.85	1.15	50.28	24.88	61.79	1,064	3,961	2,012	86.12	58
HPC-Coder-V2 + OptiML-X2026.02		7.55	1.05	49.31	25.72	60.88	1,086	4,124	2,048	85.11	58
GPT 5.1 + OptiML-X2026.02		7.9	—	48.62	26.41	60.32	1,117	4,213	2,094	84.27	58
StarCoder2 + OptiML-X2026.02		8.22	0.96	47.55	26.98	59.46	1,144	4,286	2,126	83.38	58
Qwen2.5-Coder + OptiML-X2026.02		9.1	0.87	46.37	27.91	58.24	1,183	4,368	2,172	82.46	58