Benchmark Library

Every benchmark is an independent research program: one capability claim, one task and protocol system, one leaderboard, and one evolving Data Card.

Projects: 01
Public boards: 01
Domains: 01

01Code-QA-Bench

Locate, trace, and explain implementation details in real repositories while separating code reading, memorization, and documentation use. 10 pinned SWE-Bench Python repositories GPT-5.4 LLM judge · three-axis continuous score Released