Cohere launches Embed 5 with Pro and Fast tiers for enterprise retrieval

Two tiers, one embedding space

Embed 5 Pro is optimized for quality across multimodal, multilingual, financial, code and parsed-document retrieval, according to Cohere. Embed 5 Fast is described as a lighter model for latency-sensitive, high-volume retrieval. It costs a third less than Pro while keeping the same 128K-token context, multimodal inputs, multilingual coverage and compressed output formats.

Both tiers share a single embedding space. Teams can therefore index with Pro and query with either model without rebuilding the index. Cohere points to the query path as the place where Fast matters most, because embedding latency is paid on every search and multiplied in agentic workflows that may issue dozens of searches per task.

Both models accept text, images and fused text and image inputs, cover 100+ languages, and offer output dimensions of 2048, 1536, 1024, 768, 512 and 256. Output formats are float, int8 and binary, and both support Matryoshka embeddings and self-hosting. According to Cohere, Matryoshka representations and lower-precision outputs reduce vector storage and search costs, while quantized weights lower serving requirements for private deployments.

Benchmark results reported by Cohere

Cohere says Embed 5 Pro achieves the highest average score of any model it tested across ViDoRe V3, financial documents, parsed PDFs, image retrieval and key business languages. Embed 5 is also the first model family evaluated with RCP-nDCG@10, Cohere's latest retrieval methodology, which judges retrieved documents against query-specific relevance criteria instead of a limited set of fixed labels.

• ViDoRe V3: Pro averages 85.8, a gain of 8.8 over Embed 4. That is ahead of Voyage 4 Large (83.7), Gemini Embedding 2 (83.2) and OpenAI text-embedding-3-large (75.5). Fast averages 84.5.
• Finance: Pro ranks first on FinanceBench (80.1), FinQA (90.0) and ViDoRe V3 Finance (85.0), with Fast second on each. Across these, Pro averages 3.3 points higher than Gemini Embedding 2, and its lead over OpenAI text-embedding-3-large on FinanceBench is 21.4 points.
• Parsed PDFs: Pro averages 84.8 across Cohere's suite, ahead of Voyage 4 Large (83.6), Embed 5 Fast (83.4), Gemini Embedding 2 (80.8) and Embed 4 (78.6).
• Fused text-image: Pro averages 82.3 across five datasets, against 81.2 for Fast and 61.3 for Gemini Embedding 2.
• Page images: On financial datasets Pro averages 77.0, ahead of Fast (73.2), Embed 4 (71.1), Voyage Multimodal 3.5 (70.1) and Gemini Embedding 2 (56.7).

On multilingual tests, Pro averages 77 across German, French, Spanish, Italian and Russian, compared with 76 for Voyage 4 Large and 73 for Gemini Embedding 2. Cohere says that is about 7 points better than Embed 4 on average. The company also reports gains over Embed 4 in Farsi (+13), Telugu (+12) and Hindi (+12). Its own table shows that in several languages, including Japanese, Arabic and Telugu, Gemini Embedding 2 scores higher than Pro.

Availability, pricing and what it means

Embed 5 is generally available on the Cohere API and Model Vault, Microsoft Foundry and Amazon SageMaker. Pro costs $0.12 per million tokens for text and Fast costs $0.08 per million tokens. Image input is priced at $0.40 per million tokens on both tiers.

Cohere positions Embed 5 as the retrieval foundation for search, RAG and agentic workflows, surfacing more relevant context while filtering out noise before it reaches expensive generative models. The company says this can improve answer quality and user experience while helping keep downstream inference costs under control.

Cohere recommends Pro for offline indexing and complex enterprise corpora, and Fast for interactive search, high-volume RAG and agentic retrieval. Because the two models share an index, teams can pair them without re-indexing. All benchmark figures come from Cohere's own evaluations.