Mistral AI Pricing: What's Free and Its Korean Language Limits

How the orbits differ

When to pick Mistral, and when to pick a neighboring star

A 10-minute test route in the free orbit

Frequently asked questions

Mistral AI cover image

5 things to check before you start

Mistral is a French AI company that offers both open models and a chat service

A closer look

What kind of star is Mistral, and who should board it?

The short answer: it is a star for people who cannot send their data outside, and for people who want to run models themselves. If your main job is polishing Korean documents smoothly, it drops down the priority list.

Mistral is the name used for the family of language models and the chat service built by the French AI startup Mistral AI. The company develops both open-weight models, whose weights are published so you can download and use them, and commercial models, and it also offers an option to install them on your own servers. It suits two kinds of people: corporate staff who find it hard to send internal data outside, and developers who want to run models directly or train them further. For general users who put Korean sentence quality first, it is hard to recommend as a first choice. The reasons are explained in the Korean section below.

What are the main features?

There are four main branches. They are the chat service, open models, advanced reasoning features, and a coding agent.

First, the chat service. Its former name was Le Chat, and it is described as having been renamed Vibe, an integrated work and coding agent, in May 2026. It is said to include a Work Mode that automates email, calendar, and document tasks. Second, open models. There are models by size, such as Mistral 7B, Mixtral 8x7B, Mistral Large, and Ministral, and descriptions say they can be run locally and fine-tuned, meaning further trained on your own data. Third, extended thinking and Deep Research reports, which open up on the Pro plan and above. Fourth, a coding agent mode that connects to GitHub, GitLab, Jira, Linear, and others to handle repository-level work. Image generation works by integrating Black Forest Labs' Flux family of models.

How far does the free orbit go, and how do I start?

The free tier gets you quite far. The limits are what block you. According to the summary article, the Free plan lets you use chat, access to the latest models, image generation, 500 memories, and 40+ enterprise connectors. Message and search limits are higher on Pro and above.

Monthly prices and limit figures were not in the material we gathered, so we have not listed them. Namuwiki also says the pricing explanation is very abstract, so check the limit notes directly on the official pricing page before paying. Getting started is simple: sign up for the chat service and try the Free plan. Besides the default model, there are said to be Think mode and Research mode. If you enter the same question once in Korean and once in English and compare the answers, you can gauge the Korean quality right away. Do not enter company secrets or personal information during testing.

Is it okay to use in Korean?

This is where opinions split the most. In the summary article, Korean performance that trails major competitors is listed as a drawback.

The languages the official site lists as supported are English, French, German, Italian, and Spanish, and a Korean interface (UI) is said not to be officially confirmed. The explanation is that the model does answer Korean input, but Korean is not its main language. Another summary collecting reactions from developers in Korea and abroad also reports repeated remarks about limits in reflecting subtle Korean nuance and cultural context. For work where tone matters, such as reports or blog posts, check the results with a few short texts first before deciding. On the other hand, for internal data processing or development tests where Korean quality does not matter much, this weakness carries less weight.

Can I run it on my own computer or company server, and what are the downsides?

You can run it. But the conditions differ by model. Models released as open source can be run locally for free, and descriptions say this suits places where data privacy matters. The computer specs needed differ by model size, so check the requirements before downloading. Namuwiki mentions that a small reasoning model runs on a single RTX 4090, but that is about one specific model.

At the company level, the Enterprise plan is said to offer private deployment, custom models, SAML SSO, audit logs, and more, and using it through clouds such as Amazon Web Services or Microsoft Azure is also described. A downside is the size of its ecosystem. Its plugins and apps are rated smaller than those of OpenAI and Google, so people who relied heavily on extensions for ChatGPT or Claude may find some features missing. Also, 'Mistral' is the name of the company, the models, and the service, and the service name changed from Le Chat to Vibe, so check whether older reviews are about the same service. Open weights also do not mean identical terms for every model. Some models are said to be restricted to research-purpose licenses, so if you plan to use one commercially, read each model's license separately. You can also find a basic introduction to this star on the Mistral introduction page.