Cohere Launches Embed 5 Enterprise Embedding Models in Pro and Fast Tiers
On September 30, Cohere released Embed 5, its new generation of enterprise embedding models in Pro and Fast tiers. The company claims SOTA retrieval quality and introduces a new enterprise retrieval evaluation method, RCP-nDCG@10, under which Embed 5 is the first evaluated and leading model. The launch directly targets Voyage, Google and Jina in the enterprise retrieval market.
Confirmed
- Embed 5 comes in two tiers: Pro for frontier retrieval quality and Fast for low latency and high throughput, sharing a unified embedding space for flexible switching.
- Available via Cohere API, Model Vault, Microsoft Foundry, Amazon SageMaker and multi-cloud channels in North America and beyond.
- Cohere claims Embed 5 Pro achieved the best average score among all evaluated models on ViDoRe V3, surpassing Voyage 4 Large, Gemini Embedding 2 and Jina Embeddings, supporting over 100 languages.
- Cohere says Embed 5 Fast scores more than 6 points above other fast-tier models and is priced one-third lower than Pro.
- The Cohere Embeddings team highlighted preference-aligned retrieval, stronger multilingual support, and an enterprise-security stance of keeping data on-premises.
- Embed 5 belongs to Cohere's Embed product line, complemented by Rerank models for enterprise retrieval scenarios.
Why it matters
- By proposing RCP-nDCG@10 as a new benchmark for enterprise retrieval quality, Cohere is trying to define evaluation discourse; whether the community and competitors accept it remains to be seen.
- The dual-tier, unified-embedding-space strategy balances quality and cost, and together with multi-cloud distribution, preference alignment and multilingual capability, positions Cohere to compete for the enterprise RAG and retrieval market.
2026-09-30 ~ 2026-10-01 · 6 related posts
Primary sources
- [source] Cohere launches Embed 5, a new family of enterprise embedding models — cohere · 2026-09-30
- Cohere: Embed 5 Fast beats fast-tier rivals by 6+ points at a third less cost — cohere · 2026-09-30
- [source] Cohere launches Embed 5 Pro: top average score on ViDoRe V3, trained on 100+ languages — nickfrosst · 2026-09-30
- Cohere launches Embed 5: preference-aware retrieval, stronger multilingual, big throughput gains — cohere · 2026-10-01
2 near-duplicate retellings: cohere · aidangomez