Le Chat adds Deep Research, voice and Think modes

Background

Le Chat is Mistral AI's chat assistant, a product in which a user types a question and gets an answer back. As assistants of this kind matured, a common next step was to give them modes: instead of answering only in one shot, the assistant can be asked to gather material and write it up, to think longer before replying, to handle spoken input, or to keep a set of related conversations together in one place. On 17 July 2025 Mistral AI added a group of such features to Le Chat at once ^[1].

What was added

The headline addition was Deep Research, released as a preview. It produces structured reports with references ^[1]. Mistral described it as producing lightning-fast, structured research reports on complex topics ^[1]. Two words in that description carry most of the weight. "Structured" means the output is organised as a report rather than as a chat reply. "With references" means the points made in the report are tied to sources that a reader can go and check.

Two more modes came with it. A voice mode used the Voxtral speech model ^[1]. A Think mode used the Magistral reasoning model ^[1]. The naming separates the kind of model doing the work: a speech model turns spoken audio into text and back, while a reasoning model is one that works through a problem at greater length before producing an answer.

Alongside the modes, Le Chat gained Projects, which group chats together and give each group its own library ^[1]. Mistral described Projects as helping users group related chats into focused, context-rich spaces ^[1]. The point of a library attached to a project is that the material a user has collected stays with the conversations that need it, instead of being scattered across a long single thread.

Finally, Le Chat added advanced image editing in partnership with Black Forest Labs ^[1].

Key details

Several details shape how the release should be read. Deep Research was labelled a preview, which in practice means the feature is available for people to try while it is still being refined, rather than as a finished, settled product.

The features are also deliberately different in kind. Deep Research and Think mode both aim at harder questions, but they do so in different ways: one assembles material into a written report with sources, the other spends more effort on reasoning before answering. The voice mode sits at the input and output layer, using the Voxtral speech model ^[1]. Projects work at the level of organisation rather than generation, keeping related chats and their own libraries in one space ^[1]. Image editing comes from an outside partnership rather than being built entirely in-house: Mistral worked with Black Forest Labs on it ^[1].

Why it matters

Taken together, the additions turn Le Chat from a place where you ask a question and read a reply into a workspace with several kinds of work in it: research that ends in a cited report, longer reasoning, spoken interaction, images, and a way to keep projects apart from one another. Each of these existed in some form in other assistants; what this release shows is a single product assembling them at once, and, in the case of Deep Research, presenting the output as a referenced document rather than a conversational answer ^[1].

For readers trying to follow how assistants developed, the interesting part is the shift in what counts as an answer. A structured report with references is meant to be checked rather than simply read, and a per-project library is meant to be reused rather than re-explained each time ^[1].