Cohere releases Embed 5 with Pro and Fast retrieval models
Two tiers share an embedding space, allowing different query choices without rebuilding the same index.
THE STORY IN THREE POINTS
- Embed 5 was announced on September 30.
- Pro and Fast share a compatible embedding space.
- Cohere reports text and image inputs with multilingual support.
What was announced
Cohere released Embed 5 on September 30 in Pro and Fast tiers. The models convert content into representations used for retrieval rather than generating a conversational answer. Cohere says both tiers share an embedding space, enabling teams to index with Pro and query using either tier without rebuilding the index.
The details that matter
The release describes text and image inputs, support for more than 100 languages, and a 128,000-token context window. Cohere presents Fast as an option for latency-sensitive workloads and Pro for retrieval quality. Its comparative scores are company-reported evaluations, not results independently replicated by this publication.
Editorial context
Retrieval quality is most visible when the right evidence is absent from an answer. An organization could evaluate an embedding model using questions drawn from its own documents, including cases with similar terminology but different meanings. The comparison should track which relevant passages were missed as well as how quickly results appeared. Shared representations may make experimentation easier, but the editorial question remains whether a specific collection becomes easier to search accurately. That requires a defined document set and human judgments about relevance, rather than a generic leaderboard alone.
Sources: Cohere: Introducing Embed 5—A new family of frontier embedding models ↗
