Skip to content

ExploitGym

A controlled benchmark for evaluating whether AI agents can extend vulnerability-triggering inputs into working exploits.

14 models

Higher is better.

  1. 1GPT-6 AstraOpenAI42.4
  2. 2GPT-6.1 SolOpenAI35.1
  3. 3GPT-5.6 SolOpenAI33.7
  4. 4GPT-5.6 TerraOpenAI23.2
  5. 5GPT-6 SolOpenAI22.1
  6. 6MiMo-V2.6-ProXiaomi17.8
  7. 7DeepSeek V4.1 Flash 1017DeepSeek15.3
  8. 8GLM-5.3Z.AI15.0
  9. 9GPT-5.5OpenAI13.4
  10. 10GPT-5.6 LunaOpenAI12.4
  11. 11GPT-6 LunaOpenAI11.6
  12. 12GPT-5.4OpenAI6.0
  13. 13MiMo-V2.6-FlashXiaomi6.0
  14. 14Muse Spark 1.1Meta0.8
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Working exploit generation
Tasks
898 exploitation tasks
Authors
Zhun Wang, Nico Schiller, Hongwei Li, Srijiith Sesha Narayana, Milad Nasr, Nicholas Carlini, Xiangyu Qi, Eric Wallace, Elie Bursztein, Luca Invernizzi, Kurt Thomas, Yan Shoshitaishvili, Wenbo Guo, Jingxuan He, Thorsten Holz, Dawn Song (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.