Cohere says Embed 5 can share an index across two models

The models, released on September 30, accept text and images. Cohere positions Pro for quality-critical indexing and the cheaper Fast for live searches.

The models, released on September 30, accept text and images. Cohere positions Pro for quality-critical indexing and the cheaper Fast for live searches.
Cohere announced on September 30, 2026 that Embed 5 Pro and Embed 5 Fast were generally available through the Cohere API, Model Vault, Microsoft Foundry and Amazon SageMaker. According to the company, businesses can prepare a document index with Pro and use the cheaper Fast model for searches without rebuilding that index.
An embedding model turns text or an image into a vector—a list of numbers that search software can compare to find related material. Cohere positions Embed 5 as a way to select relevant documents before passing them to models that generate answers.
Both models accept text, images and combined text-image inputs. Cohere says they can represent a page image directly or combine an image with its metadata, such as descriptive information, in one vector. This gives businesses an alternative to extracting only text from documents whose tables, charts or layout carry meaning.
Cohere lists the following text prices and input limits. Tokens are the units of text a model processes; the context window is how much it can handle at once.
| Feature | Embed 5 Pro | Embed 5 Fast |
|---|---|---|
| Price per million text tokens | $0.12 | $0.08 |
| Context window | 128K tokens | 128K tokens |
| Languages supported | More than 100 | More than 100 |
Cohere also promotes results measured with RCP-nDCG@10, its evaluation method that judges results against relevance criteria for each query. Those tests use embedding similarity scores to reorder a fixed set of candidate documents. They therefore measure reranking quality, not how well a model finds documents in the first stage of a search.
These are Cohere’s own results, not independent performance verification. The reported averages do not establish how much search quality an individual business would retain when using Fast queries on its own Pro-indexed documents.

The proposed bills would require outside validation, impose $25,000-per-violation fines and let residents sue over foreseeable harms from tools whose safeguards are bypassed.
2 min

Repeated attacks have hit several providers within days, affecting households and media and raising concerns about services that depend on connectivity.

Reported cases at Google and Rapid7 show how delegated tasks can carry hostile text—and why safeguards on network requests still matter.
4 min
2 min