Atlas

Benchmarks

← All benchmarks

ExploitGym V8 intended-vulnerability exploits

Agents · 2026-05-11

Count of ExploitGym V8 JavaScript-engine tasks exploited through the intended vulnerability with standard mitigations disabled. The official suite contains 181 V8 tasks.

Top models (higher is better)

ModelScore
GPT-5.6 Sol71.0
Claude Mythos Preview38.0
GPT-5.538.0
GPT-5.422.0
Opus 4.78.0
Opus 4.63.0
Muse Spark 1.13.0
Gemini 3.1 Pro Preview2.0
GLM-5.10.0
Loading Atlas data…