AIR-Bench Long-Doc: leaderboard

Metric: Long-document Recall@10 (%). Source: huggingface.co. 45 models tracked.

Top models

#ModelScore
1BM25 (Reranker: mmarco-mMiniLMv2-L12-H384-v1)75.01
2bge-en-icl_few-shot74.63
3BM25 (Reranker: bge-reranker-v2-m3)74.2
4BM25 (Reranker: jina-reranker-v2-base-multilingual)73.9
5NV-Embed-v273.45
6bge-en-icl_zero-shot73.43
7BM25 (Reranker: bce-reranker-base_v1)73.43
8NV-Embed-v172.08
9bge-multilingual-gemma271.41
10BM25 (Reranker: gte-multilingual-reranker-base)71.36

Interactive version: theaggregate.ai/benchmark?slug=air-bench-long-doc · How It Works · Data refreshed daily, snapshot 2026-10-09.