SSEBenchSoftware Security Benchmark for AI Code Agents

Measure how well AI coding agents fix real security vulnerabilities

🔒

Real-World Vulnerability Tasks

Every task is a publicly disclosed bug in an open-source C, Go or Rust project, paired with its upstream fix.

🐳

Isolated, reproducible environments using layered Docker images. Seamless model/agent decoupling through LiteLLM proxy.

🤖

Run Claude Code, Codex, OpenCode or your own agent against any model.

📊

Every patch is graded on build success, PoC reproduction, functional tests and the tests that came with the upstream fix.

🔧

Easy integration of new models via YAML configs and new agents via Docker. Supports sandbox and sidecar execution modes.