FinQA
We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.
- ModelScoreYour cost / mo
- 1Embed 5 Pro · Cohere90 score
- 2Embed 5 Fast · Cohere88.8 score
“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.
How each maker reported it
- Embed 5 Pro launch post
Cohere's own numbers. ViDoRe V3 scores use RCP-nDCG@10, Cohere's new metric that re-orders a fixed candidate set, so it reflects reranking quality rather than first-stage retrieval. Multilingual rows are a subset of the post's 15 languages. 'High Finance' and some sets are Cohere-internal. Cross-model retrieval: querying a Pro-indexed corpus with Fast keeps 98.4% of Pro-on-Pro nDCG@10.
- Embed 5 Fast launch post
Cohere's own numbers. ViDoRe V3 scores use RCP-nDCG@10, Cohere's new metric that re-orders a fixed candidate set, so it reflects reranking quality rather than first-stage retrieval. Multilingual rows are a subset of the post's 15 languages. 'High Finance' and some sets are Cohere-internal. Cross-model retrieval: querying a Pro-indexed corpus with Fast keeps 98.4% of Pro-on-Pro nDCG@10.
Only each model's own maker's number is shown; scores a maker reports for rival models are left out.