Skip to content

Benchmark

ExploitGym on-target exploits (kernel)

Higher scores are better.

Models measured
11
Scores recorded
15
Sources
1
IQ weight
1.39

Score distribution

20.9
Mean
20
Median
21.3
Std dev
0
10th percentile
49
90th percentile
0
Min
073
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.39 for chat models, measured over 11 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT-5.6 Solagent: Codex CLI, model note: reasoning max, eval: 6h timeout, version: v1, focus: model, reported by: OpenAI & ExploitGym Team, date: 2026-07-13· ExploitGym leaderboard data docs73
2DeepSeek V4.1 Flashagent: MopMonk Agent, eval: 8h timeout, selected subset, version: v1, focus: agent, reported by: Mopmonk AI, date: 2026-11-04· ExploitGym leaderboard data docs49
3DeepSeek V4.1 Flashagent: MopMonk Agent, eval: 8h timeout, selected subset, version: v1, focus: agent, reported by: Mopmonk AI, date: 2026-10-21· ExploitGym leaderboard data docs38
4GPT-5.6 Solagent: Codex CLI, model note: reasoning max, eval: 2h timeout cutoff, version: v1, focus: model, reported by: OpenAI & ExploitGym Team, date: 2026-07-13· ExploitGym leaderboard data docs38
5DeepSeek V4.1 Flashagent: VARAS-OneMind (智安·宇盾), eval: 6h timeout, version: v1, focus: agent, reported by: VARAS, date: 2026-09-24· ExploitGym leaderboard data docs34
6GLM 5.3agent: VARAS-OneMind (智安·宇盾), eval: 6h timeout, version: v1, focus: agent, reported by: VARAS, date: 2026-09-11· ExploitGym leaderboard data docs22
7GPT 5.5agent: Codex CLI, model note: OpenAI's default safety filters block all GPT-5.5 exploit attempts under default prompting., eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs21
8GLM 5.2agent: DoGNAVY, eval: selected subset, version: v1, focus: agent, reported by: deepsec@DARKNAVY, date: 2026-08-22· ExploitGym leaderboard data docs20
9GLM 5.3agent: DoGNAVY, eval: 10h timeout, selected subset, version: v1, focus: agent, reported by: deepsec@DARKNAVY, date: 2026-09-10· ExploitGym leaderboard data docs17
10Claude Opus 4.6agent: Claude Code, model note: Results obtained in collaboration with Anthropic., eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs1
11GPT 5.4agent: Codex CLI, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs1
12Claude Opus 4.7agent: Claude Code, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs0
13Gemini 3.1 Pro Previewagent: Gemini CLI, eval: 2h timeout, version: v0, focus: model, reported by: ExploitGym Team, date: 2026-05-13· ExploitGym leaderboard data docs0
14GLM 5.1agent: Claude Code, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs0
15Muse Spark 1.1agent: Meta Agent (off-the-shelf agent harness), model note: helpful-only versoin, eval: 4h timeout, version: v1, focus: model, reported by: Meta AI, date: 2026-07-09· ExploitGym leaderboard data docs0

Data credits: ExploitGym leaderboard (Wang et al., UC Berkeley), arXiv:2605.11086. No licence is stated for the result data; the repository's Apache-2.0 licence covers its code only, and task data keeps its upstream licences. Literal exploit counts; hidden rows omitted.