IonBin

Independent reviews. IonBin has no partnerships with the tools on this page and was not paid for these listings. Our evaluations are our own, based on public documentation and our research. Product names are trademarks of their owners and are used for identification only. How disclosure works.

Model · Mistral AI

Mistral models (Mistral AI)

Verified Oct 17, 2026 Review by Nov 16, 2026 4 sources
Short answer

Mistral AI publishes both open-weight and commercial models. The current lineup spans frontier multimodal models (Mistral Large 4 and Medium 3.5), small Apache 2.0 models (Mistral Small 4, Ministral 3), and specialists for code, OCR, speech, embeddings and moderation. Pick it when you need weights you can host yourself or a European vendor.

What it is

Mistral is a model family from Mistral AI. Its own docs split the catalog into open-weight models, which you can download and run, and commercial models, which you use through Mistral’s API. Licenses vary by model, so check the license column before you plan a self-hosted deployment. Mistral sells its assistant plans under the Vibe name, and the pricing page also refers to Le Chat.

How it works

  1. Choose a model by job: a general multimodal model for most work, a small model for cheap high-volume tasks, or a specialist (Codestral for code, OCR for documents, Voxtral for speech).
  2. Call it through Mistral’s API, or download open weights and run them on your own infrastructure or a host such as Hugging Face.
  3. Mistral also hosts some third-party models on its platform, so the model you call is not always a Mistral model. Read the model card first.

Specs and pricing (checked 2026-10-17). API use is billed per token. Mistral’s pricing FAQ gives Mistral Large as an example at $0.50 per million input tokens and $1.50 per million output tokens, without naming the version. Batch processing halves the price. The Vibe app has a free plan, with Pro at $14.99 a month and Team at $24.99 per user per month, before tax.

Benefits

  • Open weights under permissive licenses (Apache 2.0 on Mistral Small 4 and Ministral 3) make self-hosting and fine-tuning realistic.
  • Specialist models for OCR, transcription, embeddings and moderation cover common business pipelines from one vendor.
  • Small models suit high-volume classification and routing where a frontier model would be overkill.

Limits and risks

  • Short retirement windows. Mistral’s models overview lists deprecation and retirement dates for older models. OCR 4.0 was deprecated on 9/29/2026 and retired the next day. Mistral Small 3.2 was retired on 7/31/2026 and Medium 3.1 on 8/31/2026. Pin model versions and re-check the table every 30 days.
  • Licenses vary. Some flagship models use modified licenses. Read the terms before commercial self-hosting.
  • Self-hosting is a job. Running open weights yourself moves security, uptime and evaluation work onto your team.

Checked specs

Snapshot from official sources on Oct 17, 2026. Model lineups and prices change often; this page is re-verified every 30 days.

Catalog splitOpen-weight and commercial models [1]
Frontier modelsMistral Large 4 (v26.10, open-weight, multimodal); Mistral Medium 3.5 (v26.04, multimodal, agentic and coding; Modified MIT license) [1]
Small modelsMistral Small 4 (Apache 2.0); Ministral 3 at 14B, 8B and 3B (Apache 2.0) [1]
SpecialistsCodestral (code), OCR 4.1, Voxtral (speech), Mistral Embed, Mistral Moderation 2 [1]
Third-party models hostedIncludes Z.ai GLM 5.3 (1M context) [1]
API pricing (checked 2026-10-17)Per token, input and output billed separately; the pricing FAQ's example is Mistral Large at $0.50 input / $1.50 output per 1M tokens; batch 50% off; cached input up to 90% off [2]
Vibe plans (checked 2026-10-17)Free; Pro $14.99/month; Team $24.99/user/month; Enterprise by quote (prices exclude tax) [2]
MCPCustom MCP connectors (beta) in Vibe [2]
Recent retirements (checked 2026-10-17)Mistral Small 3.2: deprecated 4/30/2026, retired 7/31/2026 (use Small 4); Mistral Medium 3.1: deprecated 5/22/2026, retired 8/31/2026 (use Medium 3.5); OCR 4.0: deprecated 9/29/2026, retired 9/30/2026 (use OCR 4.1) [1]

Best-fit stacks

Stack 08

Dev, web & IT

Design, code, host and ship sites and internal tools, with an AI editor doing most of the typing.

Agent compatibility

API accessYesOfficial docs ↗
AppsVibe, web and mobile (the pricing page also refers to Le Chat)
Also offered viaOpen weights downloadable for many models

Levels come from each tool's recorded, sourced compatibility. How the levels work.

Sources

  1. Models overview · Mistral AIofficial accessed Oct 17, 2026
  2. Pricing · Mistral AIofficial accessed Oct 17, 2026
  3. Model Hub documentation · Hugging Faceofficial accessed Oct 17, 2026
  4. What are open source large language models? · IBMreference accessed Oct 17, 2026

Practitioner views are restated in our words and labeled as views, not measured results. Product facts come from official sources dated above.

Suggest an edit

IonBin's wiki isn't user-edited. Corrections go to our Directory Curator, who checks them against primary sources before anything changes. Spot something wrong or out of date on “Mistral models (Mistral AI)”? Email us with a link to a primary source.

Suggest an edit by email or write to [email protected]