Image: Mistral AI

Mistral AI has launched a Responses API in public preview, a new way for developers to build AI agents on its platform. The French company says the interface makes it easier for apps to follow each step a model takes, and it is built to match a format many developers already use.

Today, we’re introducing the Responses API in public preview, made for building agents on Mistral.

Each response returns the model’s answer, its reasoning and its tool calls as separate items, so your app can handle every step more easily than with Chat Completions.

It works with all chat models hosted by Mistral, including Large 4.

— Mistral AI (@MistralAI) October 28, 2026

What is new

Until now, most developers have talked to Mistral models through its Chat Completions API, which returns a reply as a message. The Responses API instead returns an output list in which the model's answer, its reasoning and any tool calls arrive as separate items. According to Mistral, that lets an app handle every step more easily than with Chat Completions.

The company's documentation says the API uses the OpenAI Responses format. That detail matters for teams that have already built on OpenAI's version, since Mistral says the API works well for bringing an existing Responses integration over as well as for starting fresh.

Mistral says the API works with all chat models it hosts, including Mistral Large 4. It is served in the EU and US on Mistral AI Cloud Inference.

What developers can do with it

During the preview, the docs list support for text input and output, image and file inputs, structured output in JSON mode or JSON Schema, and streaming. Developers can also call their own functions, set a reasoning effort level and view plaintext reasoning on models that support it, and count input tokens before sending a request.

One limit is worth knowing up front. The API is stateless, so the app has to keep the conversation history and send it back with every request, including the output items from earlier responses. The examples in the docs use Mistral Large 4 with a high reasoning effort setting.

Two more models on the platform

In the same thread, Mistral said it is now serving two fast, low cost models from other labs. DeepSeek V4.1 Flash is in public preview and is pitched for long context coding and agent tasks. GLM 5.3 Flash is generally available and aimed at subagents, tool calls and coding.

The announcement follows the public preview of Mistral Large 4 on October 6, which Mistral describes as an open weight multimodal model with a 1 million token context window.

Why it matters

For developers, the main appeal is portability. Code written for one provider's agent interface can more easily point at Mistral, which gives teams another option for cost, data location or vendor choice. Keeping reasoning and tool calls as separate items also makes agent behavior easier to log, debug and display.

The feature is still a preview, so behavior may change before a full release, and Mistral asks users to report problems to its support team. Pricing for the API itself was not detailed in the announcement at the time of this writing. Full parameters are in the API reference.