Skip to content

Benchmark

ExploitGym on-target exploits

Higher scores are better.

Models measured
16
Scores recorded
27
Sources
1
IQ weight
1.24

Score distribution

144.1
Mean
130
Median
104.7
Std dev
12
10th percentile
293
90th percentile
4
Min
4368
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.24 for chat models, measured over 16 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1GPT-6 Astraagent: unstated, eval: token-budgeted, version: unstated, focus: model, reported by: OpenAI, date: 2026-09-03· ExploitGym leaderboard data docs368
2GPT-6.1 Solagent: unstated, eval: token-budgeted, version: unstated, focus: model, reported by: OpenAI, date: 2026-09-29· ExploitGym leaderboard data docs305
3Claude Opus 5.5agent: unstated, eval: 6h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-09-22· ExploitGym leaderboard data docs300
4GPT-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 docs293
5Claude Opus 5.5agent: unstated, eval: 2h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-09-22· ExploitGym leaderboard data docs289
6DeepSeek 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 docs226
7GPT-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 docs216
8DeepSeek 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 docs206
9Claude Opus 5agent: unstated, eval: 6h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-07-24· ExploitGym leaderboard data docs191
10DeepSeek 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 docs186
11Claude Opus 5agent: unstated, eval: 2h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-07-24· ExploitGym leaderboard data docs171
12GLM 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 docs159
13GLM 5.3agent: DoGNAVY, eval: 10h timeout, selected subset, version: v1, focus: agent, reported by: deepsec@DARKNAVY, date: 2026-09-10· ExploitGym leaderboard data docs157
14GLM 5.3agent: Claude Code, eval: rescaled 6h timeout, version: unstated, focus: model, reported by: Z.ai, date: 2026-08-18· ExploitGym leaderboard data docs130
15GPT 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 docs129
16Claude Opus 4.8agent: unstated, eval: 6h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-05-28· ExploitGym leaderboard data docs120
17GLM 5.3agent: Claude Code, eval: rescaled 2h timeout, version: unstated, focus: model, reported by: Z.ai, date: 2026-08-18· ExploitGym leaderboard data docs105
18Claude Opus 4.8agent: unstated, eval: 2h timeout, version: unstated, focus: model, reported by: Anthropic, date: 2026-05-28· ExploitGym leaderboard data docs80
19GLM 5.2agent: DoGNAVY, eval: selected subset, version: v1, focus: agent, reported by: deepsec@DARKNAVY, date: 2026-08-22· ExploitGym leaderboard data docs79
20GPT 5.4agent: Codex CLI, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs61
21GLM 5.2agent: Claude Code, eval: rescaled 6h timeout, version: unstated, focus: model, reported by: Z.ai, date: 2026-08-18· ExploitGym leaderboard data docs39
22GLM 5.2agent: Claude Code, eval: rescaled 2h timeout, version: unstated, focus: model, reported by: Z.ai, date: 2026-08-18· ExploitGym leaderboard data docs29
23Claude 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 docs16
24Claude Opus 4.7agent: Claude Code, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs12
25Gemini 3.1 Pro Previewagent: Gemini CLI, eval: 2h timeout, version: v0, focus: model, reported by: ExploitGym Team, date: 2026-05-13· ExploitGym leaderboard data docs12
26Muse 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 docs7
27GLM 5.1agent: Claude Code, eval: 2h timeout, version: v1, focus: model, reported by: ExploitGym Team, date: 2026-06-16· ExploitGym leaderboard data docs4

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

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.