Cohere
Cohere Embed 5 lets teams index with Pro and query with Fast
Cohere priced its Embed 5 family at $0.12 and $0.08 per million tokens for text, and says both tiers share one embedding space.
Cohere released Embed 5, an embedding family in two tiers, Pro and Fast, that share one embedding space. A team can build its search index with Pro and answer queries with Fast without re-indexing. An embedding model turns text and images into lists of numbers that a search system compares.
Cohere announced the family Sept. 30, 2026, in its Introducing Embed 5 post. Both are generally available on the Cohere API and Model Vault, Microsoft Foundry and Amazon SageMaker.
Pro Costs $0.12 per Million Tokens and Fast Costs $0.08
Cohere priced text at $0.12 per million tokens for Pro and $0.08 for Fast as of Sept. 30. Images cost $0.40 per million tokens on both tiers. Cohere said Pro is tuned for maximum quality in offline indexing and complex enterprise collections. It said Fast suits interactive search, high-volume retrieval-augmented generation and agentic retrieval.
Both Tiers Take Text, Images and Fused Inputs
Cohere said both models accept text, images and fused text-image inputs, cover more than 100 languages and handle 128K tokens of context. Outputs run from 2048 down to 256 dimensions in float, int8 or binary formats, with Matryoshka embeddings. Cohere said shorter and lower-precision outputs reduce vector storage and search costs. It said quantized weights lower serving needs for private deployments. Both tiers support self-hosting.
Cohere Says Pro Posts the Top Average on Its Tests
Cohere said 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 the first family Cohere evaluated with RCP-nDCG@10, a retrieval method it described in a separate Sept. 30 post. It scores retrieved documents against query-specific relevance criteria rather than only fixed labels.
Pro Averages 85.8 on ViDoRe V3
On ViDoRe V3, scored with RCP-nDCG@10, Cohere reported Pro averages 85.8, up 8.8 from Embed 4. Cohere’s table lists Voyage 4 Large at 83.7, Gemini Embedding 2 at 83.2 and OpenAI text-embedding-3-large at 75.5. Fast averages 84.5, ahead of Gemini Embedding 2 and Voyage 4 Large.
Pro leads five of the eight domains and ties Voyage 4 Large on energy. Its biggest gains over Embed 4 were HR (up 11.4) and industrial (up 10.3).
Pro Ranks First and Fast Second on Finance Tests
Cohere said Pro ranks first and Fast second on FinanceBench, FinQA and ViDoRe V3 Finance. Pro scored 80.1, 90.0 and 85.0. Fast scored 80.0, 88.8 and 83.9.
Across those tests Pro averages 3.3 points above Gemini Embedding 2, the next non-Cohere model, Cohere said. On FinanceBench alone, Pro leads OpenAI text-embedding-3-large by 21.4 points.
Pro Also Leads on Parsed PDFs, Mixed Inputs and Languages
On Cohere’s parsed-document suite (documents parsed with Gemini 1.5 Flash), Cohere reported 84.8 for Pro, 83.6 for Voyage 4 Large, 83.4 for Fast, 80.8 for Gemini Embedding 2 and 78.6 for Embed 4.
On fused text-image retrieval, measured by nDCG@10 across five datasets, Cohere reported 82.3 for Pro, 81.2 for Fast and 61.3 for Gemini Embedding 2.
Across German, French, Spanish, Italian and Russian, Cohere said Pro averages 77, against 76 for Voyage 4 Large and 73 for Gemini Embedding 2. That is about 7 points above Embed 4.
What We Think
Fast costs a third less than Pro for text. A team can pay for Pro once to build the index and serve queries with Fast.
Cohere ran every comparison above. Its post says RCP-nDCG@10 is meant to give “a fuller view of performance on your own corpus.” We would test both tiers on that corpus first.