agents benchmark
Kernel Bench L3
Kernel Bench L3 evaluates agentic GPU kernel optimization across 50 problems. Qwen reports two metrics for this benchmark: median per-problem speedup over the PyTorch eager reference and the fraction of problems faster than torch.compile.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
Kernel Bench L3 leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 96.0% | 100.0% | 1 | C |
Kernel Bench L3 highlights
The top published results on this benchmark's own scale.
What is Kernel Bench L3?
Definition and scoring fields from the benchmark registry.
Kernel Bench L3 evaluates agentic GPU kernel optimization across 50 problems. Qwen reports two metrics for this benchmark: median per-problem speedup over the PyTorch eager reference and the fraction of problems faster than torch.compile.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Kernel Bench L3
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- kernel-bench-l3|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Kernel Bench L3.
Which model scores highest on Kernel Bench L3?
Qwen3.7 Max is currently ranked first with 96.0%.
What does Kernel Bench L3 measure?
Kernel Bench L3 evaluates agentic GPU kernel optimization across 50 problems. Qwen reports two metrics for this benchmark: median per-problem speedup over the PyTorch eager reference and the fraction of problems faster than torch.compile.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.