Code-QA-Bench - Doc-Dependent Tasks: leaderboard

Metric: Judge score (0-1; mean of accuracy, completeness and specificity; 100 documentation-dependent questions, full documented repository, GPT-5.4 judge). Source: arxiv.org. Saturation forecast: Around August 2027. 4 models tracked.

Top models

#ModelScore
1Claude Opus 4.60.95
2Kimi K2.60.95
3DeepSeek V4 Pro0.93
4Gemini 3.1 Pro (Preview)0.89

Interactive version: theaggregate.ai/benchmark?slug=code-qa-bench-doc-dependent-tasks · How It Works · Data refreshed daily, snapshot 2026-09-25.