Cohere Release Embed 5: Pro for indexing, Fast for queries—no need to rebuild the vector database

Article

According to Beating AI news, the Nvidia-backed enterprise AI company Cohere has released a new-generation embedding model Embed 5, available in Pro and Fast tiers.Embedding models convert text and images into vectors for search,RAG and Agent finding information. Both models support text, image, and mixed inputs, covering 100 languages, with a maximum context of token.

The two models share the same vector space. Enterprises can first use the higher-quality Pro to index documents, then use the faster Fast to process queries, without regenerating the entire set of vectors.Cohere Tests on 40 datasets showed that the combined performance of Pro for indexing and Fast for queries was equivalent to 98.4%of the all-Pro approach, with a gap of approximately 1.6%。

Cohere tests,Embed 5 Pro averaged ViDoRe V3 on 85.8, while Fast scored 84.5, higher than Voyage 4 Large's 83.7 and Gemini Embedding 2's 83.2. Fast is priced at token per million text 0.08 USD, while Pro is 0.12 USD.

Original article link https://m.theblockbeats.info/flash/369876

Key points

Cohere released a new-generation embedding model Embed 5, available in Pro and Fast tiers.
The original article states that both models support text, image, and mixed inputs, covering 100 languages, with a maximum context of token。
The original article states that both models share the same vector space, allowing Pro to be used for indexing and Fast for processing queries without regenerating the entire set of vectors.
Cohere Tests on 40 datasets showed that the combined performance of Pro for indexing and Fast for queries was equivalent to 98.4%, with a gap of approximately 1.6%。
Cohere In the tests,Embed 5 Pro averaged ViDoRe V3 points 85.8, while Fast scored 84.5，Voyage 4 Large scored 83.7，Gemini Embedding 2 scored 83.2。
The per-million-text pricing listed in the original article token is: Fast 0.08 USD，Pro 0.12 USD。
The original article describes Cohere as an Nvidia-backed enterprise AI company.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Embedding The model converts text or images into vectors, enabling search,RAG and Agent locating information accordingly. As described in the original article, enterprises can use the higher-quality Pro to process documents and the faster Fast to process user queries; because they share a vector space, switching query models does not require regenerating all document vectors.

Why it matters to readers

If this compatibility works in real-world business applications, enterprises can balance retrieval quality, query speed, and price, while reducing the work required to rebuild the vector database when changing query models.

For beginners, the key point is: Pro is oriented toward document ingestion, while Fast is oriented toward everyday queries; the performance, language coverage, and pricing in the article all come from reports or Cohere tests, so official materials and actual results should still be verified before selection.

Risks and unknowns

The evidence does not include Cohere an official announcement, so the models' availability, complete specifications, and pricing terms cannot be independently confirmed.
The performance data comes from Cohere tests, with no supporting independent third-party evaluations.
The complete list of the 40 datasets, testing methods, and specific operating conditions was not provided.
The limitations on the applicability of “without regenerating the entire set of vectors” across different databases, vector dimensions, and existing index configurations were not specified.

Cohere Release Embed 5: Pro for indexing, Fast for queries—no need to rebuild the vector database

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center