# How do you use Mistral's Le Chat?

Date: 2024-08-04 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/use-mistrals-le-chat/

**Answer:** Le Chat is Mistral AI's chat assistant at chat.mistral.ai: you sign in, choose a Mistral model such as the new Mistral Large 2, and chat as you would in ChatGPT, in English, French and many other languages. It works best for writing, translation, coding and reasoning tasks where you supply the facts, since it can't browse the web, and you should read its data terms before pasting anything private.

Le Chat is the conversational front door to the models of Mistral AI, the Paris-based company behind Mistral 7B and Mixtral. It opened as a beta in February, and on July 24 Mistral made its new flagship, Mistral Large 2, available to test there. If you have used ChatGPT, the interface will feel familiar: a message box, a conversation history and a menu for picking the model that answers you.

What makes it worth a look is less the interface than the model behind it and the company's European base. Mistral Large 2 is strong in many languages, including French, and Mistral's legal terms are written around the EU's data protection law, the GDPR.

## Getting started

1. Go to chat.mistral.ai and create an account or sign in.
2. Open the model menu and choose the model you want. Mistral Large 2 is the most capable option for demanding tasks.
3. Type your request. You can keep asking follow-up questions in the same conversation, and earlier chats are saved in your history.
4. To show someone a conversation, you can create a shared link, but anyone who has the link can read that chat, so share only conversations with nothing confidential in them.

One limit to remember: Mistral has said Le Chat cannot access the internet, so it can be wrong or out of date on recent events. Give it the facts and documents it needs, and check any claim that matters.

## What Mistral Large 2 brings

Mistral describes Large 2 as a 123-billion-parameter model with a 128,000-token context window, meaning it can take in a long document in one go (a token is a small piece of a word). It handles dozens of human languages, among them French, Spanish, Arabic, Hindi, Japanese and Chinese, and more than 80 programming languages. Mistral says it trained the model to be more careful, to say when it lacks enough information, and to keep answers concise, a choice you will notice in its shorter replies.

## A worked example in English and French

Bilingual work is a good test because translation is only half the job; tone and register matter too. French, for instance, distinguishes between the formal *vous* and the familiar *tu*.

```prompt
I run a small bakery in Quebec. Write a short notice, under 80 words, telling customers
that we will be closed on Monday for a family event and will reopen on Tuesday at 7 a.m.

Give me two versions:
1. English, warm and friendly.
2. French for Quebec customers, using "vous", warm but polite.
Do not translate word for word; make each version sound natural in its own language.
Then list any word in the French version that might read differently in France.
```

Read the result with a native speaker's ear, or ask someone who has one. Then try a follow-up that tests understanding rather than translation:

```prompt
Now explain, in French, in three sentences, why you chose "vous" instead of "tu",
and suggest a shorter French version for a social media post.
```

A second useful exercise is to paste a long English document, such as a policy or a set of meeting notes, and ask for a French summary in five bullet points with any uncertain terms flagged. The long context window means a document of many pages fits in one message.

> **Try it:** Give the bakery prompt, or one from your own work, to Le Chat with Mistral Large 2 and to another assistant you use. Compare the French versions line by line and note which sounds more natural to you, and why.

## Where your data goes

Mistral's privacy policy says its platform is hosted on Microsoft Azure with the data stored in Sweden, and its terms are governed by the GDPR and French law. Two other points matter for everyday use. On the free service, accepting the terms lets Mistral use your prompts and replies to improve its models; paid subscribers can opt out of that and turn on an incognito mode, though the opt-out covers only future conversations. And the terms ask users not to enter sensitive personal data or confidential business information, such as trade secrets, in their prompts. Treat Le Chat like any public chatbot: useful for drafts and explanations, not a place for client files.

## Which Mistral model fits which job?

Mistral is organizing its developer platform, la Plateforme, around four models. Mistral Large 2 is the general-purpose flagship: pick it for complex reasoning, long documents, careful multilingual writing and coding help. Mistral Nemo is the second general-purpose model. Try it for everyday questions, quick drafts and high-volume tasks, and compare its answers and API price with Large 2 on your own examples before choosing.

The two specialists do narrower jobs. Codestral is built for programming: writing, completing and explaining code. Mistral Embed turns text into embeddings, lists of numbers that capture meaning, which developers use to build search over their own documents rather than to chat.

In Le Chat itself, the practical rule is simple. Start with Mistral Large 2 when the answer needs to be right and well written, and try the other models in the menu when you want fast, short replies. For apps, the same models are available through Mistral's API and on cloud platforms such as Azure, Amazon Bedrock and Google's Vertex AI, and the Large 2 weights can be downloaded for research and non-commercial use under the Mistral Research License.

## Sources

- Large Enough: https://mistral.ai/news/mistral-large-2407/
- Legal terms and conditions (Terms of Use, Le Chat Terms of Service, Privacy Policy): https://mistral.ai/terms/
- Le Chat: https://mistral.ai/news/le-chat-mistral/