Mistral Opens a Responses API in Public Preview That Uses OpenAI's Format and Keeps No Conversation State

Mistral's new API is designed so that developers do not have to give up OpenAI's request format to move to Mistral. The company's documentation says the Responses API "lets you build with Mistral models using the OpenAI Responses format," and that it works for new applications or for "bringing an existing Responses integration to Mistral." Crypto Briefing's write-up of the launch frames the main effect as lower switching costs for developers already using that format. The catch is that Mistral has copied the format without copying OpenAI's server-side memory, so the switch is easiest for applications that already keep their own history.

What is in the preview#

Mistral announced the Responses API in public preview on Wednesday, saying each response returns the model's answer, its reasoning and its tool calls as separate items, and that it works with all chat models hosted by Mistral, including Large 4. The documentation lists text input and output, image and file input, structured output in JSON mode and JSON Schema, streaming, reasoning effort settings depending on the model, and an endpoint for counting input tokens. Function calling in the preview covers client-executed functions, meaning the developer's own code runs the tool and returns the result.

The API is stateless, according to the documentation: the application keeps the conversation history and sends it with each request, including output items from earlier responses. Mistral's code examples use mistral-large-4, the model Mistral put on a paid preview API earlier this month.

Where the compatibility stops#

OpenAI's own Responses API can hold conversation state for the developer. Its documentation describes a previous_response_id parameter that chains a new request to an earlier stored response, and a separate Conversations API that keeps a conversation as a durable object across sessions. It saves response objects for 30 days by default unless the developer sets store to false, and it shows the stateless pattern, resending history with store: false, as an alternative.

Mistral's preview supports only that stateless pattern. An application that leans on previous_response_id or OpenAI's Conversations API would need to start keeping its own history before it could point at Mistral; one that already resends history can move with fewer changes. For buyers who want prompts and outputs kept off the provider's servers, stateless operation is a feature rather than a gap, though Mistral's documentation does not pitch it that way.

Limits#

• This is a preview, and Mistral's documentation asks users to report issues to its support team; features and behaviour may change.
• Mistral's post says the API makes each step easier to handle than Chat Completions; that is the company's claim, not an independent test.
• The preview page does not give separate pricing for the Responses API.
• Mistral's documentation does not say whether stateful features will be added after the preview.

Sources#

• Mistral Docs: Responses API (preview)
• Mistral AI on X: Responses API public preview announcement (October 28, 2026)
• Crypto Briefing: Mistral opens Responses API in public preview, borrowing OpenAI's format for agent builders (October 28, 2026)
• OpenAI API docs: Conversation state
• Times of AI: Mistral Puts 1T-Parameter Large 4 on a Paid Preview API (October 6, 2026)