CodegenBench - BLAS x86-64: leaderboard
CodegenBench (2026) tests efficient parallel code generation across CPU architectures; this board covers 106 BLAS subroutines on an x86-64 Xeon, where the generated routine must compile against OpenBLAS and match the reference results. Pass@1 in percent, zero-shot; higher is better.
Metric: Pass@1 (%): share of problems whose first generated program compiles, runs and matches the reference output within tolerance; zero-shot through the official APIs with a short optimization system prompt, on the 106 BLAS subroutines (levels 1-3, several matrix shapes, strides and data types) compiled with gcc against OpenBLAS on a Xeon Platinum 8488C; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 74 | #79 |
| 2 | Claude Sonnet 4.6 | 70 | #161 |
| 3 | Claude Opus 4.6 | 67 | #102 |
| 4 | DeepSeek V4 Pro | 53 | #188 |
| 5 | DeepSeek V4 Flash | 48 | #258 |
| 6 | Qwen 3.6 Plus | 48 | #222 |
| 7 | Qwen 3.6 Flash | 40 | #424 |
| 8 | DeepSeek V3.2 | 35 | #406 |
| 9 | Qwen 3.5 Plus | 28 | #232 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=codegenbench-blas-x86-64 · How It Works · Data refreshed daily, snapshot 2026-11-13.