DecouvrIR: leaderboard
French information-retrieval leaderboard comparing sparse, dense, multi-vector, and cross-encoder retrieval models on mMARCO-fr and BSARD using recall, MRR, nDCG, and MAP metrics.
Metric: nDCG@10. Source: huggingface.co. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | multilingual-e5-large | 36.72 |
| 2 | multilingual-e5-base | 35.81 |
| 3 | Solon-embeddings-large-0.1 | 35.8 |
| 4 | multilingual-e5-small | 35.19 |
| 5 | biencoder-camembert-base-mmarcoFR | 33.7 |
| 6 | Solon-embeddings-base-0.1 | 33.5 |
| 7 | biencoder-camembert-L10-mmarcoFR | 32.5 |
| 8 | bge-m3 | 32.46 |
| 9 | biencoder-distilcamembert-mmarcoFR | 31.9 |
| 10 | biencoder-camembert-L8-mmarcoFR | 31.8 |
Interactive version: theaggregate.ai/benchmark?slug=decouvrir · How It Works · Data refreshed daily, snapshot 2026-09-05.