DataKernelBench - Pass Rate: leaderboard
Share of the 22 TPC-H queries whose LLM-optimized GPU implementation in DataKernelBench (2026) matches the reference output, for the configuration the leaderboard reports, in percent; higher is better.
Metric: Pass rate (%; share of the 22 queries whose optimized module matches the reference output, for the configuration the leaderboard reports; 22 TPC-H queries at scale factor 10 on one NVIDIA H100, each translated into a validated PyTorch TorchPlan; the model rewrites the query core or the full query in CUDA or Triton with up to several rounds of execution-guided repair; each model's best of its four framework-level configurations, as the leaderboard reports). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 100 | #161 |
| 2 | GPT-5.5 | 100 | #49 |
| 3 | Claude Opus 4.7 | 100 | #79 |
| 4 | Qwen 3.5 397B A17B | 100 | #246 |
| 5 | Claude Haiku 4.5 | 95.5 | #568 |
| 6 | DeepSeek V4 Flash | 90.9 | #258 |
| 7 | GPT-OSS-120B | 86.4 | #700 |
| 8 | MiniMax-M2.5 | 81.8 | #582 |
| 9 | Gemini 3.1 Pro (Preview) | 54.5 | #94 |
| 10 | Devstral 2 | 45.5 | #1003 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=datakernelbench-pass-rate · How It Works · Data refreshed daily, snapshot 2026-11-13.