Mistral Embed
Mistral AIMistral AI’s general-purpose text embedding endpoint, mistral-embed-2312: a 1,024-dimension text embedder with an 8k context, sold on La Plateforme at $0.10 per million tokens through /v1/embeddings and the batch API. It is distinct from codestral-embed, Mistral’s later code-specialized embedder.
It exists and it is a distinct product, but it is the oldest thing in this set: version 23.12, dated 2023-12-11, still listed as GA and Premier on Mistral’s own models page and unchanged since. Mistral has since shipped codestral-embed for code retrieval, which is a separate SKU with its own price, so mistral-embed is the general-purpose endpoint rather than the line’s frontier. The only retrieval number Mistral has ever published for it is 55.26 on MTEB retrieval, in the original La Plateforme announcement.
Openness
1 high confidence- weights
- closed(no checkpoint distributed
- data
- closed(training corpus not released or described)
- code
- closed(no training or inference implementation
- license
- proprietary(Premier-tier API model under Mistral’s terms of service
Mistral Embed is API-only: no weights, training data or code are published, and no checkpoint has shipped. That is worth stating because Mistral is widely known as an open-weights vendor for its generative models, a reputation that does not extend to its embedding endpoint.
- https://docs.mistral.ai/models/model-cards/mistral-embed-23-12 recorded 2026-09-11
The Mistral Embed model card records mistral-embed-2312, GA, Premier, version 23.12 dated 2023-12-11, an 8k context and $0.10 per million tokens, with features limited to the /v1/embeddings and /v1/batch endpoints. No weights, corpus or code.
- https://mistral.ai/models recorded 2026-09-11
Mistral's own models page lists 'Mistral Embed ... Embedding Text-to-text mistral-embed' as an API model, separately from Codestral Embed, and separately from the open-weight models it distributes.
Adoption
2 medium confidenceMistral publishes no download or call-volume figure for mistral-embed, so there is no artifact to count. It is also distributed more narrowly than any other product in this set, only through Mistral's own La Plateforme with no Bedrock, Azure or Vertex resale, and Mistral's own marketing attention has shifted to codestral-embed for retrieval.
- https://docs.mistral.ai/models/model-cards/mistral-embed-23-12 recorded 2026-09-11
The card records GA and Premier status with a single price and two endpoints; it names no cloud marketplace or third-party distribution.
- https://mistral.ai/models recorded 2026-09-11
The models page lists Mistral Embed as one API model among Mistral's catalog, with Codestral Embed carried separately and described as the state-of-the-art embedding offering.
Capability
2 high confidenceMistral-embed's one published number, 55.26 on MTEB retrieval, is far behind today's leaderboard; the model is also text-only, fixed-dimension and short-context, and Mistral has run no re-evaluation since its release. It remains sold and useful for its niche rather than unusable.
- https://docs.mistral.ai/models/model-cards/mistral-embed-23-12 recorded 2026-09-11
Version 23.12, dated 2023-12-11, context 8k, still GA; no updated benchmark is published on the card.
- https://mistral.ai/news/la-plateforme/ recorded 2026-09-11
'Mistral-embed, our embedding endpoint, serves an embedding model with a 1024 embedding dimension. ... This embedding model has been designed with retrieval capabilities in mind. It achieves a retrieval score of 55.26 on MTEB.'
Verified 2026-09-11