AIR-Bench: leaderboard
AIR-Bench 24.05 QA nDCG@10 mean over all 53 published domain-language columns. Retrieval and reranking systems are distinct; incomplete coverage has no aggregate. Collected as source evidence and excluded from LLM ranking by the existing benchmark policy.
Metric: QA nDCG@10 (%). Source: huggingface.co. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | inf-retriever-v1 (Reranker: bge-reranker-v2-m3) | 68.49 | |
| 2 | inf-retriever-v1-1.5b (Reranker: bge-reranker-v2-m3) | 66.41 | |
| 3 | inf-retriever-v1 (Reranker: jina-reranker-v2-base-multilingual) | 64.37 | |
| 4 | inf-retriever-v1-1.5b (Reranker: jina-reranker-v2-base-multilingual) | 62.61 | |
| 5 | BM25 (Reranker: bge-reranker-v2-m3) | 58.57 | |
| 6 | BM25 (Reranker: jina-reranker-v2-base-multilingual) | 56.28 | |
| 7 | inf-retriever-v1 | 54.47 | |
| 8 | bge-multilingual-gemma2 | 54.46 | |
| 9 | BM25 (Reranker: gte-multilingual-reranker-base) | 53.78 | |
| 10 | inf-retriever-v1 (Reranker: bge-reranker-large) | 52.42 |
Interactive version: theaggregate.ai/benchmark?slug=air-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.