Benchmarks / Vulnerability patching

Vulnerability patching

Write and validate fixes for security vulnerabilities in real codebases.

Part of Security.

• CyberBench v1.1by Vals AITask level · Overall accuracy (mean of PoC and Patch pass rates)

  75.4%MiMo V2.6 Flash · 23 Sep 2026

  Strong
• DARPA AI Cyber Challenge, Final Competitionby DARPA and ARPA-HProject level · Synthetic vulnerabilities found (scored round)

  43Team Atlanta · 8 Aug 2025

  Strong