You might’ve noticed that our benchmarks for Embed 5 look a bit different 👀 Embed 5 is the first model family evaluated with RCP-nDCG@10, our latest methodology that improves on assessing real-world retrieval quality.

Oct 2, 2026 · 7:44 PM UTC

6
11
99
8,830
Traditional nDCG only credits results already in a benchmark's answer key. RCP-nDCG@10 grades every result on its actual relevance, and its results are backed by human reviewers and MTEB.
2
2
9
1,196
Want to hear more? Stay tuned for our webinar on Thursday, October 8 on X Live. Our search modelling leadership will be discussing RCP-nDCG@10, Embed 5, and the future of search and retrieval. cohere.com/register/rcp-ndcg…

Webinar | The Future of Search and Retrieval | Cohere

cohere.com
5
879
Sort replies: Relevant Recent Liked
Replying to @cohere
Retrieval metrics that only reward the answer key miss real relevance. Grade every result, back it with human review, and publish the method so operators can defend search quality in production.
79
Replying to @cohere
Interesting perspective
128