A.S.E (AICGSecEval) - Quality: leaderboard

Code quality score (%) on Tencent's A.S.E (AICGSecEval, arXiv:2508.18106v3, Tables 2-3), measuring the share of repository-level CVE code completion tasks where generated diffs integrate cleanly and pass syntax and static checks.

Metric: Code quality score (%). Source: arxiv.org. Saturation forecast: Not forecast. 33 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4 (20250514)92.37#405
2Claude 3.7 Sonnet (20250219)91.58#421
3DeepSeek V3 (0324)85.87#715
4Claude Opus 4 (20250514)85.82#297
5Qwen 3 Coder 480B A35B Instruct85.16#652
6Gemini 2.5 Pro (Preview)84.04#258
7Qwen 3 235B A22B 2507 Instruct82.08#616
8Kimi K2 (Preview)79.9#307
9DeepSeek R1 052874.39#428
10Qwen-Coder-Plus-2024110673.78
11Grok 373.54#611
12Qwen 3 30B A3B 2507 Instruct72.89#1100
13GPT-4o (2024-11-20)72.46#859
14GLM-4.570.24#542
15Qwen 2.5 Coder 32B Instruct65.08#1547

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=a-s-e-aicgseceval-quality · How It Works · Data refreshed daily, snapshot 2026-11-21.