AI Radar

Mistral AI models

Mistral AI is a Paris-based AI company that says it was founded in April 2023. It develops open-weight and commercial language models, as well as models for code, OCR, speech, embeddings and moderation, and offers enterprises and governments AI products including developer tools, applications and compute. Its first model, Mistral 7B, was released in September 2023 under the Apache 2.0 licence.

Chronology

  1. 2026

    3 models
    1. Announced, release date unknown
      Mistral Large · Language, Multimodal Mistral Large 4
      Evidence incomplete Announced · 1 news item Some details lack sourcesIncomplete record
    2. Released
      Mistral Medium · Language, Multimodal, Reasoning Mistral Medium 3.5

      Mistral Medium 3.5 is a dense 128B-parameter Mistral AI model that handles instruction following, reasoning and coding in one set of weights, with text and image input and a 256k-token context…

      • Adds reasoning effort set per request (none or high); the model card says it replaces Magistral in Le Chat.
      • Released as open weights under a Modified MIT licence; Mistral listed Medium 3.1 as Premier (commercial) with no weights link.
      Primary source Available · 1 news item
    3. Released
      Mistral Small · Language, Multimodal, Reasoning · Milestone Mistral Small 4

      Mistral Small 4 is a mixture-of-experts model from Mistral AI with 119B total parameters, released on 16 March 2026 as Apache 2.0 open weights and on Mistral's API. Mistral describes it as its first…

      • Adds reasoning effort set per request (none or high) in the same model, instead of a separate Magistral reasoning model.
      • Moves to a mixture-of-experts design with 128 experts, 4 active per token, and 119B total parameters.
      Primary source Available · 1 news item

  2. 2025

    13 models
    1. Released
      Ministral · Language, Multimodal, Reasoning · Milestone Ministral 3

      Ministral 3 is a series of dense Mistral AI language models in 3B, 8B and 14B sizes aimed at edge and local use, released on 2 December 2025 with Mistral Large 3. Each size has base, instruct and…

      • Adds a 14B size alongside 3B and 8B.
      • Each size comes with a separate reasoning checkpoint next to the base and instruct checkpoints.
      Primary source Available · no news linked
    2. Released
      Mistral Large · Language, Multimodal · Milestone Mistral Large 3

      Mistral Large 3 is a sparse mixture-of-experts language model from Mistral AI with 675B total and 41B active parameters, released on 2 December 2025 with the Ministral 3 series as part of Mistral 3…

      • Uses a sparse mixture-of-experts architecture with 41B of 675B parameters active; Mistral calls it its first mixture-of-experts model since Mixtral.
      • Has a context window of 256k tokens.
      Primary source Available · no news linked
    3. Released
      Magistral · Language, Multimodal, Reasoning Magistral Medium 1.2

      Magistral Medium 1.2 is a reasoning model in Mistral AI's Magistral line that takes text and image input and returns text, released on 17 September 2025 together with Magistral Small 1.2 and offered…

      Primary source Retired · no news linked
    4. Released
      Magistral · Language, Multimodal, Reasoning Magistral Small 1.2

      Magistral Small 1.2 is a 24B-parameter reasoning model from Mistral AI, built on Mistral Small 3.2 and released on 17 September 2025 on Mistral's API and as Apache 2.0 open weights. It accepts text…

      • Adds a vision encoder, so the model accepts image input and can reason over images as well as text.
      • Reasoning content is wrapped in new [THINK] and [/THINK] special tokens, and the reasoning prompt is given in the system prompt.
      Primary source Available · no news linked
    5. Released
      Mistral Medium · Language, Multimodal Mistral Medium 3.1

      Mistral Medium 3.1 is a multimodal language model from Mistral AI that takes text and image input and returns text, released on 12 August 2025 on Mistral's API under the commercial Premier licence…

      • Mistral says the update improves tone and performance.
      Primary source Retired · no news linked
    6. Released
      Magistral · Language, Reasoning Magistral Medium 1.1

      Magistral Medium 1.1 is a reasoning model in Mistral AI's Magistral line, released on 24 July 2025 together with Magistral Small 1.1 and offered as API model magistral-medium-2507. The docs list…

      Primary source Retired · no news linked
    7. Released
      Magistral · Language, Reasoning Magistral Small 1.1

      Magistral Small 1.1 is an update of Magistral Small 1.0, Mistral AI's 24B-parameter reasoning model, released on 24 July 2025 as open weights under Apache 2.0 and as API model magistral-small-2507…

      • Better tone and behaviour, with improved LaTeX and Markdown formatting and shorter answers to easy prompts.
      • Less likely to enter infinite generation loops.
      Primary source Available · no news linked
    8. Released
      Mistral Small · Language, Multimodal Mistral Small 3.2

      Mistral Small 3.2 is a minor update of Mistral Small 3.1 Instruct, a 24B-parameter model with text and image input, published as open weights under Apache 2.0 and offered as API model…

      • Mistral AI says it follows precise instructions better.
      • Produces fewer infinite or repetitive generations, according to Mistral AI.
      Primary source Available · no news linked
    9. Released
      Magistral · Language, Reasoning · Milestone Magistral Medium

      Magistral Medium is the enterprise variant of Magistral, which Mistral AI calls its first reasoning model, trained with reinforcement learning on top of Mistral Medium 3. It launched as a preview in…

      • Adds step-by-step reasoning, trained with reinforcement learning alone on top of Mistral Medium 3.
      • The docs list a 40k context window, against 128k for Mistral Medium 3.
      Primary source Retired · no news linked
    10. Released
      Magistral · Language, Reasoning Magistral Small

      Magistral Small is the 24B-parameter open-weights variant of Magistral, which Mistral AI calls its first reasoning model, released under Apache 2.0 and also offered as an API model. Its model card…

      • Adds long reasoning chains before the answer, through fine-tuning on Magistral Medium traces and reinforcement learning on top of Mistral Small 3.1.
      Primary source Available · no news linked
    11. Released
      Mistral Medium · Language, Multimodal · Milestone Mistral Medium 3

      Mistral Medium 3 is a commercial multimodal model from Mistral AI with text and image input, offered through its API, on Amazon SageMaker and as the model behind Le Chat Enterprise. Mistral AI…

      • Accepts image input in addition to text.
      • Has a context window of 128k tokens.
      Primary source Retired · no news linked
    12. Released
      Mistral Small · Language, Multimodal Mistral Small 3.1

      Mistral Small 3.1 is a 24B-parameter model from Mistral AI that builds on Mistral Small 3, adding image understanding and a context window of up to 128k tokens. It was released with base and…

      • Adds image input, so the model can understand images as well as text.
      • Context window of up to 128k tokens, up from 32k for Mistral Small 3.
      Primary source Available · no news linked
    13. Released
      Mistral Small · Language · Milestone Mistral Small 3

      Mistral Small 3 is a 24B-parameter language model from Mistral AI, released with pretrained and instruction-tuned open weights under Apache 2.0 and offered via API on la Plateforme and through…

      • Released under Apache 2.0; Mistral AI said it would move away from MRL-licensed models for its general-purpose models.
      • Has 24B parameters, published as a pretrained and an instruction-tuned checkpoint.
      Primary source Available · no news linked

  3. 2024

    14 models
    1. Released
      Mistral Large · Language Mistral Large 24.11

      Mistral Large 24.11 is an update of Mistral AI's Mistral Large 2 language model, a 123B dense model released alongside Pixtral Large; the changelog calls it Mistral Large 2.1. Mistral describes…

      • Mistral reports better long-context understanding.
      • Mistral reports more accurate function calling.
      Primary source Available · no news linked
    2. Released
      Pixtral · Language, Multimodal Pixtral Large

      Pixtral Large is a 124B multimodal model from Mistral AI that takes text and images and outputs text, built on Mistral Large 2 with a 123B decoder and a 1B-parameter vision encoder. Mistral calls it…

      • Grows to 124B parameters in total: a 123B multimodal decoder and a 1B-parameter vision encoder.
      • Its multimodal decoder is built on Mistral Large 2 instead of Mistral Nemo.
      Primary source Available · no news linked
    3. Released
      Ministral · Language Ministral

      Les Ministraux are two language models, Ministral 3B and Ministral 8B, that Mistral AI introduced for on-device and edge use on the first anniversary of Mistral 7B, both with a context length of up…

      Primary source Available · no news linked
    4. Released
      Mistral Small · Language Mistral Small v24.09

      Mistral Small v24.09 is a 22-billion-parameter language model from Mistral AI, an upgrade of Mistral Small v24.02 that Mistral positions between Mistral NeMo and Mistral Large 2. It was offered on…

      • Weights were published for download under the Mistral Research License, for non-commercial self-deployment.
      • Mistral reports improvements in human alignment, reasoning and code over Mistral Small v24.02.
      Primary source Available · no news linked
    5. Released
      Pixtral · Language, Multimodal · Milestone Pixtral 12B

      Pixtral 12B is a multimodal model from Mistral AI that takes images and text and outputs text, combining a new 400M-parameter vision encoder with a 12B decoder based on Mistral Nemo. It was released…

      • Adds image input through a new 400M-parameter vision encoder trained from scratch; images are processed at their native resolution and aspect ratio.
      • Pairs that encoder with a 12B multimodal decoder built on Mistral Nemo; Mistral says it keeps text-only capabilities and can replace Mistral Nemo 12B as a drop-in.
      Primary source Available · no news linked
    6. Released
      Mistral Large · Language · Milestone Mistral Large 2

      Mistral Large 2 is the second generation of Mistral AI's Mistral Large language model, with 123 billion parameters and a 128k-token context window. It was offered on la Plateforme as…

      • Instruct weights were published for download under the Mistral Research License, for research and non-commercial use.
      • Mistral says it trained the model on a very large proportion of code, drawing on its work on Codestral 22B and Codestral Mamba, and on a large share of multilingual data.
      Primary source Available · no news linked
    7. Released
      Mistral · Language Mistral NeMo

      Mistral NeMo is a 12B-parameter language model that Mistral AI built with NVIDIA, released with base and instruct weights under Apache 2.0 and offered on la Plateforme and as an NVIDIA NIM. It has a…

      • Has 12B parameters, up from 7B in Mistral 7B, while keeping a standard architecture that Mistral says lets it replace Mistral 7B as a drop-in.
      • Uses a new tokenizer, Tekken, based on Tiktoken, in place of the SentencePiece tokenizer of earlier Mistral models.
      Primary source Available · no news linked
    8. Released
      Language, Code Codestral Mamba

      Codestral Mamba is a 7B code model from Mistral AI based on the Mamba architecture rather than a Transformer, which Mistral says gives linear-time inference. It was released under Apache 2.0 with…

      Primary source Available · no news linked
    9. Released
      Language Mathstral

      Mathstral is a 7B language model from Mistral AI, built on Mistral 7B and specialised in mathematics and STEM subjects, aimed at problems that need complex, multi-step logical reasoning. Mistral…

      • Specialised for mathematics and STEM subjects; Mistral compares its results by subject with those of Mistral 7B.
      Primary source Available · no news linked
    10. Released
      Mistral · Language Mistral 7B v0.3

      Mistral 7B v0.3 is a minor update to Mistral AI's open 7B language model, released on 22 May 2024 as base and instruct checkpoints with open weights on Hugging Face under Apache 2.0 and served on…

      • Vocabulary extended to 32768 tokens.
      • The instruct model adds function calling, together with support for the v3 tokenizer.
      Primary source Available · no news linked
    11. Released
      Mixtral · Language Mixtral 8x22B

      Mixtral 8x22B is a sparse mixture-of-experts language model from Mistral AI with 141B total and 39B active parameters, released on 17 April 2024 with an instruct version, open weights under Apache…

      • Larger sparse mixture-of-experts model: 141B total and 39B active parameters, against 46.7B and 12.9B for Mixtral 8x7B.
      • Context window of 64K tokens, up from 32k for Mixtral 8x7B.
      Primary source Available · no news linked
    12. Released
      Mistral Large · Language · Milestone Mistral Large

      Mistral Large, later named Mistral Large 1.0, is a language model that Mistral AI called its new flagship when it was released on 26 February 2024 on la Plateforme, on Microsoft Azure and in the le…

      Primary source Retired · no news linked
    13. Released
      Mistral · Language Mistral Next

      Mistral Next was a prototype language model that Mistral AI offered in the le Chat beta from 26 February 2024, next to Mistral Large and Mistral Small, and described as designed to be brief and…

      Primary source Retired · no news linked
    14. Released
      Mistral Small · Language Mistral Small

      Mistral Small, later named Mistral Small 1.0, is a language model that Mistral AI released on 26 February 2024 alongside Mistral Large, through the mistral-small-2402 API endpoint and in le Chat…

      Primary source Retired · no news linked

  4. 2023

    4 models
    1. Released
      Mistral · Language Mistral 7B v0.2

      Mistral 7B v0.2 is an update to Mistral AI's open 7B language model, first offered on 11 December 2023 as Mistral 7B Instruct v0.2 behind the mistral-tiny API endpoint of la Plateforme, later…

      • Context window raised to 32k tokens, from 8k in v0.1.
      • Sliding-window attention dropped and rope-theta set to 1e6.
      Primary source Available · no news linked
    2. Released
      Mistral Medium · Language · Milestone Mistral Medium

      Mistral Medium, later named Mistral Medium 1.0, was the model behind the mistral-medium endpoint at the beta launch of Mistral AI's API platform, la Plateforme, on 11 December 2023. Mistral AI…

      Primary source Retired · no news linked
    3. Released
      Mixtral · Language · Milestone Mixtral 8x7B

      Mixtral 8x7B is a sparse mixture-of-experts language model from Mistral AI with 46.7B parameters, 12.9B of them used per token, released on 11 December 2023 with open weights under Apache 2.0 and an…

      • Same architecture as Mistral 7B except that each layer has eight feedforward experts, of which a router uses two per token.
      Primary source Available · no news linked
    4. Released
      Mistral · Language · Milestone Mistral 7B

      Mistral 7B is the first open model from Mistral AI: a 7.3-billion-parameter language model released on 27 September 2023 with open weights under Apache 2.0, together with a chat fine-tune, Mistral…

      Primary source Available · no news linked

Lineage

  1. Mistral 7B
    Variants and derived
    Idefics2 (Hugging Face) · derived: other; Zephyr 7B Alpha (Hugging Face) · derived: fine tune; Zephyr 7B β (Hugging Face) · derived: fine tune; Mistral 7B v0.2 · revision; PPLX Online LLMs (Perplexity) · derived: fine tune
  2. Mistral 7B v0.2
    Based on
    Mistral 7B · revision
    Variants and derived
    Mistral 7B v0.3 · revision
  3. Mistral Medium
  4. Mixtral 8x7B
    Successors
    Mixtral 8x22B
  5. Mistral Large
    Successors
    Mistral Large 2
  6. Mistral Next

    No recorded relations.

  7. Mistral Small
  8. Mixtral 8x22B
    Predecessors
    Mixtral 8x7B
    Variants and derived
    Zephyr 141B-A35B (Hugging Face) · derived: fine tune
  9. Mistral 7B v0.3
    Successors
    Mistral NeMo
    Based on
    Mistral 7B v0.2 · revision
    Variants and derived
    Mathstral · derived: other
  10. Codestral Mamba

    No recorded relations.

  11. Mathstral
    Based on
    Mistral 7B v0.3 · derived: other
  12. Mistral NeMo
    Predecessors
    Mistral 7B v0.3
    Variants and derived
    Pixtral 12B · derived: other
  13. Mistral Large 2
    Predecessors
    Mistral Large
    Variants and derived
    Mistral Large 24.11 · revision; Pixtral Large · derived: other
  14. Mistral Small v24.09
    Predecessors
    Mistral Small
    Successors
    Mistral Small 3
  15. Pixtral 12B
    Successors
    Pixtral Large
    Based on
    Mistral NeMo · derived: other
  16. Ministral
    Successors
    Ministral 3
  17. Mistral Large 24.11
    Successors
    Mistral Large 3
    Based on
    Mistral Large 2 · revision
  18. Pixtral Large
    Predecessors
    Pixtral 12B
    Based on
    Mistral Large 2 · derived: other
  19. Mistral Small 3
  20. Mistral Small 3.1
    Predecessors
    Mistral Small 3
    Variants and derived
    Magistral Small · derived: fine tune; Magistral Small 1.1 · derived: fine tune; Ministral 3 · derived: distillation
  21. Mistral Medium 3
    Predecessors
    Mistral Medium
    Variants and derived
    Magistral Medium · derived: fine tune
  22. Magistral Medium
    Based on
    Mistral Medium 3 · derived: fine tune
    Variants and derived
    Magistral Small · variant
  23. Magistral Small
    Based on
    Magistral Medium · variant; Mistral Small 3.1 · derived: fine tune
  24. Mistral Small 3.2
    Predecessors
    Mistral Small 3.1
    Successors
    Mistral Small 4
    Variants and derived
    Magistral Small 1.2 · derived: fine tune
  25. Magistral Medium 1.1
    Predecessors
    Magistral Medium
  26. Magistral Small 1.1
    Predecessors
    Magistral Small
    Based on
    Mistral Small 3.1 · derived: fine tune
  27. Mistral Medium 3.1
    Predecessors
    Mistral Medium 3
  28. Magistral Medium 1.2
  29. Magistral Small 1.2
    Based on
    Mistral Small 3.2 · derived: fine tune
  30. Ministral 3
    Predecessors
    Ministral
    Based on
    Mistral Small 3.1 · derived: distillation
  31. Mistral Large 3
  32. Mistral Small 4
    Predecessors
    Mistral Small 3.2
  33. Mistral Medium 3.5
    Predecessors
    Mistral Medium 3.1
  34. Mistral Large 4

    No recorded relations.

Families

Not in a family

About the organisation

  • Description Mistral AI is a Paris-based AI company that says it was founded in April 2023. It develops open-weight and commercial language models, as well as models for code, OCR, speech, embeddings and moderation, and offers enterprises and governments AI products including developer tools, applications and compute. Its first model, Mistral 7B, was released in September 2023 under the Apache 2.0 licence. Primary source[1] [2] [3] [4]

Company · https://mistral.ai · 34 models listed here

Sources

  1. [1] Legal notice | Mistral

    Other · Mistral AI

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://mistral.ai/legal/ States the legal form (simplified joint stock company), the Paris Trade and Companies Register number and the registered office in Paris, France. The notice names the company Mistral. Living page; content as seen on 2026-10-01.

  2. [2] About Mistral | Open, frontier AI for all.

    Other · Mistral AI

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://mistral.ai/about/ Company page with the mission, founding in April 2023, key dates and the three co-founders. Living page; content as seen on 2026-10-01. https://mistral.ai/about redirects (301) here.

  3. [3] Models Overview

    Documentation · Mistral AI

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://docs.mistral.ai/models Living page; content as seen on 2026-10-01. Its Deprecated section (Deprecated and retired models) lists model, version, API id, deprecation and retirement dates in M/D/YYYY format, and the alternative. HTML title Mistral Docs. The older address https://docs.mistral.ai/getting-started/models/models_overview/ redirects here (two 308 redirects).

  4. [4] Mistral 7B

    Announcement · Mistral AI · published

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://mistral.ai/news/announcing-mistral-7b/ HTML title and heading both read Mistral 7B. The address without the trailing slash (linked from the docs) redirects (301) here.