supermarket

Benchmark

ExploitBench

Measures: How far an agent climbs a ladder of exploitation steps, from reaching vulnerable code to arbitrary code execution, against 41 patched bugs in Chromium's V8 engine with mitigations on.

Maintainer
Carnegie Mellon University
Introduced
2026
Status
Active
Saturation
Headroom
Typical provenance
Independent
In scoring set
No

PaperOfficial leaderboard

Leaderboard

Show independent results only

Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.

Leaderboard
ModelMean capabilityProvenanceSourceEvaluated
GPT-5.5
OpenAI · Unknown effort
47.4%Source not statedEpoch AI via Epoch AI—
GPT-5.5
OpenAI · Unknown effort
41.8%Source not statedEpoch AI via Epoch AI—
GPT-5.5
OpenAI · Unknown effort
41.6%Source not statedEpoch AI via Epoch AI—
GPT-5.5
OpenAI · Unknown effort
32.6%Source not statedEpoch AI via Epoch AI—
Claude Opus 4.7
Anthropic
26.5%Source not statedEpoch AI via Epoch AI—
Gemini 3.1 Pro
Google DeepMind
26.1%Source not statedEpoch AI via Epoch AI—
Claude Opus 4.7
Anthropic
23.9%Source not statedEpoch AI via Epoch AI—
Claude Sonnet 4.6
Anthropic
23.6%Source not statedEpoch AI via Epoch AI—
Claude Sonnet 4.6
Anthropic
22.7%Source not statedEpoch AI via Epoch AI—
Kimi K2.6
Moonshot AI
18.4%Source not statedEpoch AI via Epoch AI—
GLM-5.1
Z.ai (Zhipu AI)
18.1%Source not statedEpoch AI via Epoch AI—
Kimi K2.6
Moonshot AI
16.3%Source not statedEpoch AI via Epoch AI—
GLM-5.1
Z.ai (Zhipu AI)
16.0%Source not statedEpoch AI via Epoch AI—
Gemini 3.1 Pro
Google DeepMind
15.7%Source not statedEpoch AI via Epoch AI—
Claude Haiku 4.5
Anthropic · Unknown effort
13.7%Source not statedEpoch AI via Epoch AI—
Claude Haiku 4.5
Anthropic · Unknown effort
13.4%Source not statedEpoch AI via Epoch AI—
MiniMax-M2.7
MiniMax
13.3%Source not statedEpoch AI via Epoch AI—
MiniMax-M2.7
MiniMax
13.3%Source not statedEpoch AI via Epoch AI—
  • Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0.

Sources