Skip to content
Weights Not Walls
Menu

Cybench

Reported

Capture-the-flag security tasks taken from professional competitions.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://cybench.github.io/
Score shown
Epoch's column "Unguided % Solved", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
Results
22 in all: 3 for open-weight models, 19 for closed ones. 3 linked to a model page here.

Best open-weight result: meta-llama/Llama-3.1-405B-Instruct, 7.5%, number 20 of 22 results.

Cybench results
#ModelWeightsSettingScoreDateSource
1claude-opus-4-6_unknownAnthropicclosedeffort unknown93.0%not datedOpus 4.6 System Card
2claude-opus-4-5-20251101_unknownAnthropicclosedeffort unknown82.0%not datedOpus 4.5 System Card
3claude-sonnet-4-5-20250929_unknownAnthropicclosedeffort unknown60.0%not datedOpus 4.5 System Card
4claude-sonnet-4-5-20250929Anthropiccloseddefault55.0%not datedCybench leaderboard
5grok-4-0709xAIcloseddefault43.0%not datedGrok 4 Model Card
6claude-opus-4-1-20250805Anthropiccloseddefault42.0%not datedCybench leaderboard
7grok-4-1xAIcloseddefault39.0%not datedGrok 4.1 Model Card
8claude-opus-4-20250514Anthropiccloseddefault38.0%not datedCybench leaderboard
9claude-sonnet-4-20250514Anthropiccloseddefault35.0%not datedCybench leaderboard
10grok-4-fastxAIcloseddefault30.0%not datedGrok 4 Fast Model Card
11o3-mini-2025-01-31_mediumOpenAIclosedeffort medium22.5%not datedCybench leaderboard
12claude-3-7-sonnet-20250219Anthropiccloseddefault20.0%not datedCybench leaderboard
13claude-3-5-sonnet-20240620Anthropiccloseddefault17.5%not datedCybench leaderboard
14gpt-4.5-preview-2025-02-27OpenAIcloseddefault17.5%not datedCybench leaderboard
15gpt-4o-2024-11-20OpenAIcloseddefault12.5%not datedCybench leaderboard
16claude-3-opus-20240229Anthropiccloseddefault10.0%not datedCybench leaderboard
17o1-mini-2024-09-12_mediumOpenAIclosedeffort medium10.0%not datedCybench leaderboard
18o1-preview-2024-09-12OpenAIcloseddefault10.0%not datedCybench leaderboard
19gemini-1.5-pro-001-feb24Google DeepMindcloseddefault7.5%not datedCybench leaderboard
20meta-llama/Llama-3.1-405B-InstructEpoch: Llama-3.1-405B-Instruct, Meta AIopendefault7.5%not datedCybench leaderboard
21mistralai/Mixtral-8x22B-Instruct-v0.1Epoch: Mixtral-8x22B-Instruct-v0.1, Mistral AIopendefault7.5%not datedCybench leaderboard
22meta-llama/Meta-Llama-3-70B-InstructEpoch: Meta-Llama-3-70B-Instruct, Meta AIopendefault5.0%not datedCybench leaderboard

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