A.S.E (AICGSecEval): leaderboard

Repository-level AI-generated code security benchmark from Tencent Wukong Code Security Team (arXiv:2508.18106v3, Tables 2-3) evaluating 37 Code LLMs on real GitHub CVE patches with dual mutations. Scored by weighted Overall score (0.6*Security + 0.3*Quality + 0.1*Stability) on a 0-100 scale.

Metric: Overall weighted score (0-100). Source: arxiv.org. Saturation forecast: Not forecast. 33 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude 3.7 Sonnet (20250219)63.01#421
2Qwen 3 235B A22B 2507 Instruct60.15#616
3Qwen 3 Coder 480B A35B Instruct59.31#652
4DeepSeek V3 (0324)58.59#715
5Claude Sonnet 4 (20250514)57.14#405
6Qwen 3 30B A3B 2507 Instruct56.59#1100
7Kimi K2 (Preview)55.29#307
8GPT-4o (2024-11-20)55.1#859
9Qwen-Coder-Plus-2024110653.55
10Claude Opus 4 (20250514)52.71#297
11Grok 352.18#611
12DeepSeek R1 052851.76#428
13Gemini 2.5 Pro (Preview)51.02#258
14GLM-4.549.8#542
15Qwen 2.5 Coder 7B Instruct45.6#3914

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 · How It Works · Data refreshed daily, snapshot 2026-11-21.