Skip to content
Weights Not Walls
Menu

ExploitBench

Reported

No description yet: we only describe an eval set when we can do it correctly.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Score shown
Epoch's column "Mean capability", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
Results
20 in all: 6 for open-weight models, 14 for closed ones. 6 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K2.6, 18.4%, number 12 of 20 results.

ExploitBench results
#ModelWeightsSettingScoreDateSource
1claude-mythos-preview-earlyAnthropiccloseddefault73.8%not datedvia Epoch AI
2claude-mythos-preview-earlyAnthropiccloseddefault70.6%not datedvia Epoch AI
3gpt-5.5_unknownOpenAIclosedeffort unknown47.4%not datedvia Epoch AI
4gpt-5.5_unknownOpenAIclosedeffort unknown41.8%not datedvia Epoch AI
5gpt-5.5_unknownOpenAIclosedeffort unknown; Harness: Codex41.6%not datedvia Epoch AI
6gpt-5.5_unknownOpenAIclosedeffort unknown; Harness: Codex32.6%not datedvia Epoch AI
7claude-opus-4-7Anthropiccloseddefault26.5%not datedvia Epoch AI
8gemini-3.1-pro-previewGoogle DeepMindcloseddefault26.1%not datedvia Epoch AI
9claude-opus-4-7Anthropiccloseddefault23.9%not datedvia Epoch AI
10claude-sonnet-4-6Anthropiccloseddefault23.6%not datedvia Epoch AI
11claude-sonnet-4-6Anthropiccloseddefault22.7%not datedvia Epoch AI
12moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshotopendefault18.4%not datedvia Epoch AI
13zai-org/GLM-5.1Epoch: glm-5.1, Z.ai (Zhipu AI)opendefault18.1%not datedvia Epoch AI
14moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshotopendefault16.3%not datedvia Epoch AI
15zai-org/GLM-5.1Epoch: glm-5.1, Z.ai (Zhipu AI)opendefault16.0%not datedvia Epoch AI
16gemini-3.1-pro-previewGoogle DeepMindcloseddefault15.7%not datedvia Epoch AI
17claude-haiku-4-5-20251001_unknownAnthropicclosedeffort unknown13.7%not datedvia Epoch AI
18claude-haiku-4-5-20251001_unknownAnthropicclosedeffort unknown13.4%not datedvia Epoch AI
19MiniMaxAI/MiniMax-M2.7Epoch: MiniMax-M2.7, MiniMaxopendefault13.3%not datedvia Epoch AI
20MiniMaxAI/MiniMax-M2.7Epoch: MiniMax-M2.7, MiniMaxopendefault13.3%not datedvia Epoch AI

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method