Generate README Eval: leaderboard

Benchmark for generating structured README files from entire GitHub repositories, evaluating long-context codebase summarization with BLEU, ROUGE, semantic similarity, structure, information retrieval, code consistency, and readability metrics.

Metric: Score (self-reported). Source: benchmarklist.com. Status: saturated. 14 models tracked.

Top models

#ModelScore
1GPT-4o (2024-08-06)33.13
2O1 Mini (2024-09-12)33.05
3GPT-4o Mini (2024-07-18)32.16
4Mistral Nemo Instruct (2407)25.62
5Llama 3.1 8B Instruct24.43

Interactive version: theaggregate.ai/benchmark?slug=generate-readme-eval · How It Works · Data refreshed daily, snapshot 2026-09-05.