benchgap
Agentic · tools

CWE-bench leaderboard

As of 2026-11-18, the highest measured score on CWE-bench is 75.0% by Claude Fable 5.2. 68 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra75.5%estimated ± 6.2 pp, low confidence
2Claude Fable 5.275.0%measured
3Gemini 3.7 Flash72.2%estimated ± 6.2 pp, low confidence
4Gemini 4 Carbon71.5%estimated ± 5.0 pp, medium confidence
5GPT-6 Cyber71.3%estimated ± 9.0 pp, low confidence
6Claude Fable 5.169.0%estimated ± 6.2 pp, low confidence
7Gemini 4 Hydrogen68.1%measured
8Claude Opus 5.567.1%measured
9Muse Wonder Full66.6%estimated ± 10.3 pp, low confidence
10Muse Wonder66.1%measured
11Gemini 4 Argon65.0%measured
12Claude Opus 564.2%estimated ± 6.2 pp, low confidence
13GPT-6.2 Sol64.2%estimated ± 6.2 pp, low confidence
14GPT-6 Astra64.2%measured
15Hy464.1%measured
16Claude Mythos 5.264.1%estimated ± 9.0 pp, low confidence
17GPT-6.1 Sol63.0%estimated ± 4.7 pp, medium confidence
18ERNIE 5.5 Flash Preview62.9%estimated ± 11.0 pp, low confidence
19Atria Dawn Preview62.5%estimated ± 11.0 pp, low confidence
20Ling 3.1 Flash62.4%estimated ± 11.0 pp, low confidence
21Hy4 flash62.4%estimated ± 11.0 pp, low confidence
22LongCat-2.562.3%estimated ± 11.0 pp, low confidence
23Ultramarine 1 Flash62.2%estimated ± 11.0 pp, low confidence
24Inkling 2 Small60.9%estimated ± 6.2 pp, low confidence
25LFM3-1.7B60.0%estimated ± 11.0 pp, low confidence
26Claude Sonnet 5.560.0%estimated ± 4.7 pp, medium confidence
27GPT-5.6 Cyber59.9%estimated ± 9.0 pp, low confidence
28GPT-5.559.3%estimated ± 6.2 pp, low confidence
29Claude Fable 557.8%estimated ± 6.2 pp, low confidence
30DeepSeek V4.1 Flash 101757.6%measured
31Ripple-65B Preview56.9%estimated ± 11.0 pp, low confidence
32GPT-5.6 Sol56.1%estimated ± 6.2 pp, low confidence
33Grok 4.655.3%estimated ± 6.6 pp, low confidence
34Gemini 3.1 Pro54.9%estimated ± 10.3 pp, low confidence
35Grok 4.754.7%measured
36Claude Sonnet 554.5%estimated ± 6.2 pp, low confidence
37Qwen4 Max54.5%estimated ± 6.2 pp, low confidence
38Claude Haiku 4.554.1%estimated ± 10.3 pp, low confidence
39Claude Sonnet 4.654.1%estimated ± 10.3 pp, low confidence
40DeepSeek V4 Flash 073154.1%estimated ± 10.3 pp, low confidence
41Gemini 3.6 Flash54.1%estimated ± 10.3 pp, low confidence
42GLM-5.254.1%estimated ± 10.3 pp, low confidence
43GPT-5.3-Codex-Spark54.1%estimated ± 10.3 pp, low confidence
44Grok 4.2054.1%estimated ± 10.3 pp, low confidence
45Llama 4 Maverick54.1%estimated ± 10.3 pp, low confidence
46MiMo-V2.5-Pro54.1%estimated ± 10.3 pp, low confidence
47MiniMax M354.1%estimated ± 10.3 pp, low confidence
48Muse Ember54.1%estimated ± 10.3 pp, low confidence
49Muse Spark 1.254.1%estimated ± 10.3 pp, low confidence
50Muse Spark Open54.1%estimated ± 10.3 pp, low confidence
51Nemotron 3 Ultra54.1%estimated ± 10.3 pp, low confidence
52Qwen3.8-27B54.1%estimated ± 10.3 pp, low confidence
53Claude Mythos 5.153.9%estimated ± 9.0 pp, low confidence
54Muse Spark 1.353.9%estimated ± 6.6 pp, low confidence
55GPT-6.1 Terra53.8%estimated ± 5.0 pp, medium confidence
56GPT-5.6 Terra53.7%estimated ± 9.0 pp, low confidence
57Claude Opus 4.852.8%estimated ± 6.2 pp, low confidence
58Gemini 3.5 Flash52.8%estimated ± 6.2 pp, low confidence
59GPT-6 Sol52.4%estimated ± 4.7 pp, medium confidence
60DeepSeek V4 Pro51.7%estimated ± 11.0 pp, low confidence
61Claude Opus 4.751.3%estimated ± 6.2 pp, low confidence
62Claude Mythos Preview49.8%estimated ± 9.0 pp, low confidence
63Kimi K348.0%estimated ± 4.7 pp, medium confidence
64Hy4 preview47.3%measured
65Gemini 3.8 Flash47.0%estimated ± 4.7 pp, medium confidence
66GPT-5.6 Luna46.2%estimated ± 9.0 pp, low confidence
67Claude Haiku 5.545.9%estimated ± 4.7 pp, medium confidence
68Mistral Large 445.6%measured
69Qwen3.8 Max45.4%measured
70Kimi K3.1 Pro45.3%measured
71LFM2.5-24B-A2B44.9%estimated ± 11.0 pp, low confidence
72MiMo-V2.6-Pro44.0%estimated ± 5.0 pp, medium confidence
73GLM-5.3-Flash43.9%estimated ± 4.7 pp, low confidence
74Beam 1.0 501B-A23B43.1%estimated ± 10.5 pp, low confidence
75GPT-6 Luna42.9%estimated ± 4.7 pp, low confidence
76GLM-5.342.0%measured
77Step 5 Preview41.8%estimated ± 4.7 pp, low confidence
78GPT-5.441.8%estimated ± 9.0 pp, low confidence
79MiMo-V2.6-Flash41.8%estimated ± 9.0 pp, low confidence
80MiniMax M3.1 Flash38.6%estimated ± 5.0 pp, low confidence
81Muse Spark 1.138.2%estimated ± 9.0 pp, low confidence
82Ling 3.0 Flash29.6%estimated ± 4.7 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General