Mistral AI: Model Weights, APIs and History

Mistral AI develops downloadable models, hosted APIs and enterprise AI. Its chronology distinguishes licenses, public previews and weight releases.

kindCompany
foundedApr 2023

Latest

Updated

Models with several routes to deployment

Mistral AI develops language models and software for using them through hosted services and customer deployments. Its about page dates its beginning to April 2023. A recurring part of its work is releasing downloadable models, alongside APIs and enterprise products. The access terms need to be checked for each release.

The September 2023 Mistral 7B release supplied weights under Apache 2.0. In December 2025 the Mistral 3 family extended that approach to Large 3 and smaller Ministral models. Public checkpoints let developers operate and adapt those models outside Mistral's hosted API, subject to their stated terms.

A preview is a distinct release stage

On October 6, 2026, Mistral opened a public API preview of Large 4. The announcement promised weights later in October. It therefore establishes hosted preview access on that date, rather than a completed weight release. Calling a hosted model and downloading a checkpoint offer different control over deployment and adaptation.

Mistral's benchmark comparisons describe its own evaluation results. They help explain the intended workloads, but do not independently establish that a release is best for every application. Its history includes both shipped artifacts and staged access, with those distinctions preserved in the linked events.

Sources

  1. [1]
    About Mistral | Open, frontier AI for all.

    Mistral AI · retrieved Oct 23, 2026

  2. [2]
    Mistral 7B

    Mistral AI · Sep 27, 2023

  3. [3]
    Introducing Mistral 3

    Mistral AI · Dec 2, 2025

  4. [4]
    Introducing Mistral Large 4

    Mistral AI · Oct 6, 2026

Mistral AI history

2026

│
Mistral releases the weights of Large 4

Mistral released downloadable weights of Large 4 after its October API preview, and set API pricing from that day at $0.70 per million input tokens and $2.10 per million output tokens. The checkpoint uses a modified MIT license that requires separate commercial permission for Model-as-a-Service businesses.

Mistral AI
Model
│
Mistral opens a public preview of Mistral Large 4

Mistral's largest model, a 1.05 trillion parameter mixture-of-experts with 52 billion active, went on its API at $1.36 input and $4.18 output per million tokens, with weights promised by the end of October. Its reinforcement learning run was still going when the preview opened.

Mistral AI
Model
■
Mistral raises €3 billion at a €21 billion valuation

A Samsung-led Series D, the largest equity round any European technology company has raised, arrived three years after Mistral's launch and nearly doubled its valuation from the year before. The company says the money goes toward owning data centers rather than just renting capacity.

Mistral AI
Funding

2025

■
Mistral ships its third family under Apache 2.0

Mistral Large 3, a 675 billion parameter mixture of experts with 41 billion active, alongside Ministral models at 14B, 8B and 3B in base, instruct and reasoning variants. Mistral says Large 3 entered LMArena at number two among open non-reasoning models.

Mistral AI
Model
│
Mistral ships Magistral, its first reasoning model

Magistral Small is a 24 billion parameter open-weight model released under Apache 2.0; Magistral Medium is Mistral's enterprise reasoning tier. Both aim at transparent, checkable multi-step reasoning across professional domains and languages.

Mistral AI
Model
│
Mistral releases Small 3, a 24B open model tuned for speed

Mistral AI released Small 3, a 24 billion parameter model under Apache 2.0 that it said matched larger models like Llama 3.3 70B while running more than three times faster on comparable hardware. It reported 81 percent accuracy on MMLU at 150 tokens per second.

Mistral AI
Model

2024

│
Mistral releases Codestral under a non-production license

With LangChain, LlamaIndex, Continue, Tabnine

A 22 billion parameter code model covering 80 languages, downloadable but not for commercial use without a separate deal. The new Mistral AI Non-Production License was the news: the company that made its name on Apache 2.0 weights had found the line.

Mistral AI
Model
│
Frontier labs sign the Seoul safety commitments

With OpenAI, Anthropic, Google, Meta, Microsoft, Amazon, xAI, Cohere, G42, IBM, Inflection AI, Naver, Samsung Electronics, Technology Innovation Institute, Zhipu AI

At the UK and South Korea's Seoul summit, sixteen frontier labs including OpenAI, Anthropic, Google, Meta and Zhipu AI committed to publish safety frameworks with risk thresholds and to hold back models that breach them. Voluntary, and the signatory list kept growing afterwards.

Department for Science, Innovation and Technology
Policy

2023

│
Mistral 7B beats models twice its size

An Apache-licensed 7B model from a four-month-old Paris company that outperformed 13B competitors. Europe got a frontier lab, and small models stopped being a consolation prize.

Mistral AI
Model