General

ExploitBench Internal Port leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-11-05, the highest measured score on ExploitBench Internal Port is 42.0% by GPT-6 Cyber. 11 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 4 Carbon	49.5%	estimated ± 4.9 pp, low confidence
2	GPT-6 Cyber	42.0%	measured
3	GPT-6 Astra	39.0%	measured
4	Gemini 4 Argon	28.9%	estimated ± 4.9 pp, medium confidence
5	GPT-6.1 Terra	24.5%	measured
6	GPT-6.1 Sol	21.5%	measured
7	Claude Mythos 5	9.4%	estimated ± 9.6 pp, low confidence
8	Kimi K3.1 Pro	7.4%	estimated ± 9.6 pp, low confidence
9	GPT-5.6 Sol	5.5%	measured
10	GPT-6 Sol	5.5%	measured
11	Claude Mythos Preview	4.7%	estimated ± 9.6 pp, low confidence
12	MiMo-V2.6-Pro	4.2%	estimated ± 4.9 pp, medium confidence
13	DeepSeek V4.1 Flash 1017	2.0%	estimated ± 4.9 pp, medium confidence
14	GLM-5.3	1.2%	estimated ± 9.6 pp, low confidence
15	GPT-5.6 Terra	1.0%	estimated ± 9.6 pp, low confidence
16	GPT-5.5	0.0%	estimated ± 4.9 pp, medium confidence
17	GPT-6 Luna	0.0%	measured
18	MiMo-V2.6-Flash	0.0%	estimated ± 4.9 pp, medium confidence

ExploitBench Internal Port leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General