CodegenBench - LeetKunpeng: leaderboard

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 20 LeetKunpeng kernels (the LeetSunway problems targeting Kunpeng hardware features, clang); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 9 models tracked.

Top models

#ModelScore
1Claude Opus 4.764
2Claude Opus 4.658
3Claude Sonnet 4.649
4DeepSeek V4 Pro21
5Qwen 3.6 Plus14
6DeepSeek V4 Flash11
7Qwen 3.5 Plus5
8DeepSeek V3.20
9Qwen 3.6 Flash0

Interactive version: theaggregate.ai/benchmark?slug=codegenbench-leetkunpeng · How It Works · Data refreshed daily, snapshot 2026-09-29.