Skip to content
← Models
ClosedMistral AI/10 trustVerified

mistral-embed

Mistral's 1024-dim retrieval embedding for teams that want to stay inside the European AI stack.

Visit official page →

About

mistral-embed (also listed as Mistral Embed 2312) is Mistral AI's general-purpose text embedding model, released in December 2023 as the retrieval counterpart to its chat lineup. It outputs fixed 1024-dimensional vectors, accepts up to 8,192 tokens per input, and is positioned squarely at semantic search and RAG pipelines rather than code, where Codestral Embed takes over.

A 2026 third-party audit of the live API measured 55.26 on MTEB retrieval tasks and confirmed the model rejects any dimension-reduction parameter, so 1024 dims is what you store. At $0.10 per million tokens with EU hosting options, it is the natural embedding choice for teams already generating with Mistral models.

Specs

Context window

8,192 tokens per input

Pricing

$0.10 per 1M tokens

Free quota

Limited free trial credits on new La Plateforme accounts

API access

Mistral La Plateforme API (embeddings endpoint); also via Azure, Bedrock, and Vertex marketplaces

Weights

Closed — API only

Released

Not disclosed

Benchmarks

  • MTEB retrieval tasks: 55.26 (third-party measurement against the live API, 2026; Vercel AI Gateway model page)
  • No provider-published MTEB headline found

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

semantic-searchrageu-hosted-searchdocument-retrieval

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: 1024-dim output, 8,192-token input limit, and $0.10/1M pricing cross-checked across OpenRouter docs, Vercel AI Gateway, and a 2026 live-API audit. Strengths: stable since Dec 2023, strong fit for EU-residency requirements, pairs natively with Mistral generation models. Docked for: (1) weights closed; (2) the 55.26 MTEB figure is third-party, not provider-published; (3) no dimension control.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.0/10
Maintenance · recency, releases, issue load
8.0/10
Openness · license + docs
3.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 1024 dims, 8,192-token inputs, $0.10/1M cross-checked across OpenRouter, Vercel AI Gateway, and a 2026 live-API audit
  • · 55.26 MTEB retrieval figure is third-party (Vercel AI Gateway), not provider-published
  • · Released 2023-12-11 as Mistral Embed 2312; no dimension-reduction parameter

Sources