Benchmark

CWE-bench v1

Remediating real vulnerabilities: the model has to find and patch the flaw, not just describe it.

Score history →

Model scores

#ModelScore
1Gemini 4 Argon68%