Embed Multilingual 3
Embed Multilingual 3 supports retrieval across languages through Cohere's API. The official repository describes multilingual training and provides its tokenizer and integration guidance.
Model behavior
The model card describes training with English and non-English pairs across more than 100 languages. It gives examples of separate document and query encoding. Local private deployment is a commercial access route, rather than a public weight download.Embed Multilingual 3 tokenizer and API guide
How to use it well
Practical guidance. Use a multilingual evaluation set that includes questions and documents in different languages. A good result for translated English questions does not establish quality for locally written questions, names or abbreviations. Keep the original text available with the result so a fluent reviewer can check meaning. Compare the current model family before committing a new archive to this older vector space.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | embed-multilingual-v3.0 |
| Context window | 512 tokens Text input token limit for this v3 variant. Cohere Embed specifications |
| Vector dimensions | 1024Cohere Embed specifications |
| Language scope | More than 100 languagesEmbed Multilingual 3 tokenizer and API guide |
Related models
- Embed 5 Fast
Cohere · Embed
Low-latency embedding model that shares its vector space with Embed 5 Pro.
Launch date not confirmedAPI - Embed Multilingual Light 3
Cohere · Embed
Compact multilingual text embedding model in the v3 family.
Launch date not confirmedAPI