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

Code security 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 code reduces tailored static-analysis vulnerability alerts.

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

Top models

#ModelScoreOverall rank
1Qwen 3 235B A22B 2507 Instruct48.03#616
2Claude 3.7 Sonnet (20250219)46.72#421
3GPT-4o (2024-11-20)45.65#859
4Qwen 3 30B A3B 2507 Instruct45.46#1100
5Qwen 3 Coder 480B A35B Instruct42.69#652
6DeepSeek V3 (0324)40.89#715
7Qwen 2.5 Coder 7B Instruct40.78#3914
8Grok 338.64#611
9DeepSeek R1 052838.01#428
10Qwen-Coder-Plus-2024110637.98
11Kimi K2 (Preview)37.82#307
12GLM-4.535.92#542
13Claude Sonnet 4 (20250514)34.78#405
14Qwen 3 4B 2507 (Thinking)33.57#1433
15Qwen 2.5 Coder 14B Instruct32.24#1832

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