benchgap
Agentic · tools

ExploitGym leaderboard

As of 2026-11-21, the highest measured score on ExploitGym is 53.5% by Gemini 4 Carbon. 123 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Gemini 4 Carbon53.5%measured
2GPT-6 Cyber48.5%measured
3Claude Fable 5.244.6%estimated ± 6.3 pp, medium confidence
4GPT-6 Astra42.4%measured
5GPT-6.2 Sol41.1%estimated ± 6.3 pp, medium confidence
6Gemini 4 Neon40.1%estimated ± 6.3 pp, medium confidence
7GPT-6.2 Astra40.1%estimated ± 6.3 pp, medium confidence
8Gemini 4 Argon39.5%measured
9GPT-6.1 Terra39.1%estimated ± 6.3 pp, medium confidence
10Muse Wonder Full38.5%estimated ± 2.7 pp, high confidence
11Claude Mythos 5.238.1%measured
12GPT-6.1 Sol35.1%measured
13Atria Dawn Preview33.8%estimated ± 9.2 pp, low confidence
14GPT-5.6 Sol33.7%measured
15Claude Opus 5.533.3%measured
16Qwen4 Max33.1%estimated ± 9.8 pp, low confidence
17Claude Opus 532.7%estimated ± 2.7 pp, high confidence
18LongCat-2.532.3%estimated ± 9.2 pp, low confidence
19GPT-5.6 Cyber32.1%measured
20Ling 3.1 Flash31.7%estimated ± 9.2 pp, low confidence
21Claude Fable 530.8%estimated ± 2.7 pp, high confidence
22Claude Fable 5.130.6%estimated ± 2.7 pp, high confidence
23Claude Sonnet 5.528.6%estimated ± 2.7 pp, high confidence
24Inkling 2 Small28.0%estimated ± 9.2 pp, low confidence
25GPT-5.5 Pro27.8%estimated ± 9.2 pp, low confidence
26GPT-5.4 Pro25.8%estimated ± 9.2 pp, low confidence
27Qwen3.8 Max Preview25.6%estimated ± 10.6 pp, low confidence
28Hy423.5%estimated ± 10.5 pp, low confidence
29Claude Mythos 5.123.4%measured
30GPT-5.6 Terra23.2%measured
31Claude Mythos 522.5%estimated ± 9.2 pp, low confidence
32GPT-6 Sol22.1%measured
33ERNIE 5.5 Flash Preview21.0%estimated ± 9.2 pp, low confidence
34Kimi K3.1 Pro20.7%measured
35Gemini 3.8 Flash19.5%estimated ± 2.7 pp, high confidence
36Ornith-1.5-397B19.0%estimated ± 9.2 pp, low confidence
37Fugu Cyber17.8%estimated ± 9.3 pp, low confidence
38MiMo-V2.6-Pro17.8%measured
39Gemini 3.8 Flash Cyber17.8%estimated ± 9.3 pp, low confidence
40Claude Mythos Preview17.5%measured
41Gemini 3.5 Flash Cyber17.5%estimated ± 9.3 pp, low confidence
42Gemini 4 Hydrogen17.0%estimated ± 6.3 pp, medium confidence
43Grok 4.716.8%estimated ± 2.7 pp, high confidence
44Hy4 preview16.4%estimated ± 5.5 pp, low confidence
45Claude Haiku 5.515.8%estimated ± 2.7 pp, high confidence
46Grok 4.615.8%estimated ± 2.7 pp, high confidence
47Muse Wonder15.4%estimated ± 2.7 pp, high confidence
48DeepSeek V4.1 Flash 101715.3%measured
49GLM-5.315.0%measured
50Kimi K315.0%estimated ± 2.7 pp, high confidence
51Grok 4.514.7%estimated ± 10.6 pp, low confidence
52Muse Spark 1.314.5%estimated ± 2.7 pp, high confidence
53Gemini 3.7 Flash13.5%estimated ± 2.7 pp, high confidence
54GPT-5.513.4%measured
55GPT-5.6 Luna12.4%measured
56Muse Spark 1.212.3%estimated ± 2.7 pp, high confidence
57Claude Sonnet 512.2%estimated ± 2.7 pp, high confidence
58Claude Opus 4.812.1%estimated ± 2.7 pp, high confidence
59Claude Haiku 4.511.9%estimated ± 6.3 pp, medium confidence
60Claude Sonnet 4.611.9%estimated ± 6.3 pp, medium confidence
61GLM-5.3-Flash11.9%estimated ± 6.3 pp, medium confidence
62GPT-5.3-Codex-Spark11.9%estimated ± 6.3 pp, medium confidence
63Grok 4.2011.9%estimated ± 6.3 pp, medium confidence
64Llama 4 Maverick11.9%estimated ± 6.3 pp, medium confidence
65Muse Ember11.9%estimated ± 6.3 pp, medium confidence
66Qwen3.8 Max11.9%estimated ± 6.3 pp, medium confidence
67DeepSeek V4 Flash 073111.9%estimated ± 2.7 pp, high confidence
68Qwen3.8-27B11.7%estimated ± 2.7 pp, high confidence
69Claude Opus 4.611.7%estimated ± 9.2 pp, low confidence
70GPT-6 Luna11.6%measured
71Muse Spark Open11.5%estimated ± 2.7 pp, high confidence
72GLM-5.211.2%estimated ± 2.7 pp, high confidence
73Qwen3.8-Flash-Next11.0%estimated ± 10.4 pp, low confidence
74DeepSeek V4 Pro11.0%estimated ± 9.2 pp, low confidence
75Gemini 3.5 Flash10.9%estimated ± 2.7 pp, high confidence
76dots3-note Preview10.7%estimated ± 9.2 pp, low confidence
77Kimi K2.610.4%estimated ± 9.2 pp, low confidence
78Gemini 3.6 Flash10.2%estimated ± 2.7 pp, high confidence
79MiniMax M39.7%estimated ± 2.7 pp, medium confidence
80Mistral Large 49.7%estimated ± 7.2 pp, low confidence
81Step 5 Preview9.2%estimated ± 6.4 pp, low confidence
82Nemotron 3 Ultra8.8%estimated ± 2.7 pp, medium confidence
83MiMo-V2.5-Pro8.3%estimated ± 2.7 pp, medium confidence
84Inkling7.6%estimated ± 5.5 pp, low confidence
85MiniMax M3.1 Flash6.7%estimated ± 7.2 pp, low confidence
86Sage-A1.56.7%estimated ± 9.2 pp, low confidence
87Gemini 3.1 Pro6.6%estimated ± 2.7 pp, medium confidence
88Hy3 Preview6.3%estimated ± 10.6 pp, low confidence
89GPT-5.46.0%measured
90MiMo-V2.6-Flash6.0%measured
91Apodex 1.15.2%estimated ± 10.6 pp, low confidence
92Apodex 1.1 Mini5.2%estimated ± 10.6 pp, low confidence
93Quasar 438B5.0%estimated ± 10.6 pp, low confidence
94Ling 3.0 Flash VL4.2%estimated ± 10.6 pp, low confidence
95Ling 3.0 Flash3.6%estimated ± 6.4 pp, low confidence
96Qwen3.7 Max3.2%estimated ± 10.6 pp, low confidence
97Ling 3.0 Flash Sante2.9%estimated ± 9.2 pp, low confidence
98Solar Pro 42.7%estimated ± 10.6 pp, low confidence
99Grok 4.32.2%estimated ± 10.6 pp, low confidence
100Hy31.8%estimated ± 10.6 pp, low confidence
101Qwen3.7 Plus1.7%estimated ± 10.4 pp, low confidence
102GLM-5.11.5%estimated ± 9.3 pp, low confidence
103Claude Opus 4.51.5%estimated ± 9.3 pp, low confidence
104Muse Spark1.5%estimated ± 9.3 pp, low confidence
105GLM-51.5%estimated ± 9.3 pp, low confidence
106Kimi K2.7 Code1.4%estimated ± 10.6 pp, low confidence
107GLM-4.71.1%estimated ± 10.6 pp, low confidence
108GPT-5.4 mini1.0%estimated ± 10.6 pp, low confidence
109MiniMax M2.71.0%estimated ± 10.6 pp, low confidence
110Muse Spark 1.10.8%measured
111Qwen3.6 Plus0.8%estimated ± 10.6 pp, low confidence
112Qwen3.6-27B0.8%estimated ± 10.6 pp, low confidence
113Gemini 3.5 Flash-Lite0.7%estimated ± 10.6 pp, low confidence
114Nemotron 3.5 Bolt 74B-A9B0.7%estimated ± 10.6 pp, low confidence
115Claude Opus 4.70.6%estimated ± 9.2 pp, low confidence
116A.X K20.5%estimated ± 10.6 pp, low confidence
117GPT-5.4 nano0.5%estimated ± 10.6 pp, low confidence
118GPT-50.3%estimated ± 10.6 pp, low confidence
119Qwen3.6-35B-A3B0.2%estimated ± 10.6 pp, low confidence
120Kimi K2.50.1%estimated ± 10.6 pp, low confidence
121GPT-5.10.1%estimated ± 10.6 pp, low confidence
122Qwen3.5-122B-A10B0.1%estimated ± 10.6 pp, low confidence
123K-EXAONE 2.00.1%estimated ± 10.6 pp, low confidence
124Muse Glimmer 30B0.0%estimated ± 10.6 pp, low confidence
125Mistral Medium 3.5 128B0.0%estimated ± 10.6 pp, low confidence
126MiniCPM5-2B0.0%estimated ± 10.6 pp, low confidence
127Nemotron 3.5 Lightning 30B A3B NVFP40.0%estimated ± 10.6 pp, low confidence
128MiMo-V2-Flash0.0%estimated ± 10.6 pp, low confidence
129Gemma 4 31B0.0%estimated ± 10.6 pp, low confidence
130GPT-OSS 120B0.0%estimated ± 10.6 pp, low confidence
131Granite 4.2 30B0.0%estimated ± 10.6 pp, low confidence
132Gemma 4 26B A4B0.0%estimated ± 10.6 pp, low confidence
133Ling 3.0 Tiny0.0%estimated ± 10.6 pp, low confidence
134Agents-A10.0%estimated ± 9.2 pp, low confidence
135Command A+0.0%estimated ± 12.2 pp, low confidence
136Gemini 2.5 Pro0.0%estimated ± 12.2 pp, low confidence
137Granite 4.2 8B0.0%estimated ± 12.2 pp, low confidence
138Inkling 2 Mini0.0%estimated ± 9.2 pp, low confidence
139Inkling-Small0.0%estimated ± 9.2 pp, low confidence
140Mistral Large 30.0%estimated ± 12.2 pp, low confidence
141Nemotron 3.5 Super VL 120B A12B0.0%estimated ± 9.2 pp, low confidence
142Nemotron 3.5 Super VL 120B A12B NVFP40.0%estimated ± 9.2 pp, low confidence
143Beam 1.0 501B-A23B0.0%estimated ± 9.2 pp, low confidence
144Beam 1.0 501B-A23B FP80.0%estimated ± 9.2 pp, low confidence
145Beam 1.0 501B-A23B NVFP40.0%estimated ± 9.2 pp, low confidence
146Step 3.7 Flash0.0%estimated ± 9.2 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General