Mistral

Mistral AI is a Paris company that makes open-weight and commercial AI models; its flagship, Mistral Large 4, is a 1-trillion-parameter open-weight model released in Oct 2026. It also makes the Vibe assistant (formerly Le Chat) and raised €3 billion in Sep 2026.

Mistral AI is a French AI company, founded in Apr 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix, that publishes most of its models as open weights. Its flagship is Mistral Large 4, previewed on Oct 6, 2026 and released as open weights later that month: about 1 trillion parameters with 52 billion active, 1 million tokens of context, and a modified MIT license. Its API launched at $0.70 per million input tokens and $2.10 per million output tokens. Mistral's consumer assistant, Le Chat, has been called Vibe since May 28, 2026; see Le Chat.

Pick Mistral for strong open models you can host yourself in Europe or on your own servers, for EU data handling, or for small models (Ministral 3, Mistral Small 4) under Apache 2.0. For the very top of the leaderboards, closed models from OpenAI, Anthropic and Google still lead: see AI Leaderboards. For other open models compare Qwen, DeepSeek, Gemma and gpt-oss.

The Mistral logo and the words #1 in Sovereign AI over orange and red stripes Mistral AI

Quick facts

Maker	Mistral AI, Paris, France
Founded	Apr 2023
Get it	Mistral Studio (API), Hugging Face, the Vibe app
Flagship	Mistral Large 4 (Oct 2026), nicknamed "Le Chonk"
Open weights	Yes for most models: Apache 2.0 for Small 4 and Ministral 3, the Mistral Ecosystem Open-Weight License for Large 4
API price	Large 4 from $0.70 in, $2.10 out per million tokens at launch
Funding	€3 billion Series D, valued above €21 billion (Sep 8, 2026)

Models

Model	Size	License	Notes
Mistral Large 4	~1 trillion total, 52 billion active (MoE)	Mistral Ecosystem Open-Weight License	Flagship, multimodal, 1M context
Mistral Medium 3.5	128 billion, dense	Not published	Powers Vibe's remote agents
Mistral Small 4	119 billion total, 6.5 billion active (MoE)	Apache 2.0	Includes the former Devstral and Magistral abilities
Ministral 3	3B, 8B and 14B	Apache 2.0	Small models for laptops and devices
Codestral	Not published	Commercial	Code completion, version 25.08
Voxtral 3	Mini models 4 billion	Mixed	Speech to text and text to speech
OCR 4.1	Not published	Commercial	Document reading

Sources: Mistral models page and Mistral docs model list, read Nov 6, 2026. Mistral's docs mark Medium 3.5 as deprecated, to retire Apr 30, 2027.

Mistral Large 4 is Mistral's largest model. Mistral's own launch figures include 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4, and it called it second of five models in a blind human coding test (Mistral, Oct 6, 2026). The open weights, on Hugging Face, use 128 experts with 4 active per token and run with vLLM or SGLang across eight GPUs. Its license allows commercial use and changes, but a company selling the model as a hosted service needs a separate license from Mistral.

Mistral Small 4 and Ministral 3 are the models most people can run themselves; the Ministral 3 sizes fit a laptop, and both are under Apache 2.0. See Run AI Locally.

0:29 Voxtral 3 is Mistral's speech family. In Mistral's Oct 21, 2026 launch video, a voice agent listened with Voxtral Transcribe, which also detected when each speaker had finished, reasoned with Mistral Large and answered with the latest Voxtral speech model, all three streaming at once. It handled an interruption, told the two speakers apart in real time, answered in English, French and German as asked, and ran a web search; Mistral said it targets replies in under a second.

What happened to Codestral and Devstral?

Codestral, Mistral's code completion model, is still sold; the current version is 25.08, released at the end of Jul 2025, per Mistral's docs. The Devstral coding models and the Magistral reasoning models are all marked deprecated there: Mistral folded their abilities into Mistral Small 4 (models page). Mistral's coding agent is Vibe, in the terminal (Vibe CLI), in VS Code and JetBrains, and on the web and mobile (pricing page).

Vibe (formerly Le Chat)

Mistral's assistant launched as Le Chat and was renamed Vibe on May 28, 2026, merged with the coding agent of the same name. It is free with limits, Pro is $14.99 a month and Team $24.99 a user a month (as of Nov 6, 2026). Full details are on the Le Chat page.

Funding and partners

On Sep 8, 2026 Mistral announced a €3 billion Series D led by Samsung Electronics, with EQT's Scaleup Europe Fund and PSG Equity co-leading, at a valuation above €21 billion; Mistral called it the largest equity round ever raised by a European tech company. Investors also include ASML, which led its Series C, and Nvidia. In Sep 2026 Mistral announced work with Mozilla to bring its AI into the Firefox browser.

Mistral vs the alternatives

Llama: Meta's open models, more restrictive license and no new release since Apr 2025.
Qwen and DeepSeek: Chinese open models; DeepSeek's MIT license is simpler, and its API is cheaper. Pick Mistral if you want a European maker and EU hosting.
Gemma: Google's small open models; pick Gemma or Ministral for phones and laptops.
gpt-oss: OpenAI's open-weight reasoning models under Apache 2.0.

Frequently asked questions

Is Mistral open source? Most of Mistral's models have open weights: Mistral Small 4 and Ministral 3 are under Apache 2.0, and Mistral Large 4 under a modified MIT license that requires a separate license only for selling it as a hosted service. Some models, such as Codestral and the OCR models, are offered only through Mistral's API.

What is the best Mistral model? Mistral Large 4, released in Oct 2026, is Mistral's most capable model: about 1 trillion parameters with 52 billion active and 1 million tokens of context.

How much does the Mistral API cost? Mistral Large 4 launched at $0.70 per million input tokens and $2.10 per million output tokens; Mistral's pricing page lists other models in its Studio, as of Nov 6, 2026.

Is Le Chat still available? Mistral renamed Le Chat to Vibe on May 28, 2026; it is the same assistant, free with limits or $14.99 a month for Pro.

Where is Mistral AI based? Mistral AI is based in Paris, France, and was founded in Apr 2023.