Dossier
Gemini 3.1 Pro
Coverage of Gemini 3.1 Pro in the Nexus archive.
- N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?
N-Day-Bench evaluates frontier LLMs' ability to detect security vulnerabilities in real GitHub codebases using monthly updated test cases. The benchmark involves sandboxed environments, diverse repositories with over 10k stars, and scores models like GPT-5.4 and Claude Opus 4.6 on their vulnerability discovery capabilities.