Mistral opens Responses API in public preview, borrowing OpenAI’s format for agent builders

Mistral AI released its Responses API in public preview in October 2026, a stateless endpoint at /v1/responses on Mistral's Studio/API platform that mirrors OpenAI's Responses format so developers can port existing agent workloads with lower switching costs. The API supports multimodal inputs (text, files, and images via URL or base64), JSON outputs, streaming, client-executed function calling, and model-dependent reasoning parameters, with mistral-large-4 as the primary model in its examples and official Python and TypeScript SDKs. Because the API is stateless, applications must store and resend conversation history with every request, and Mistral cautions that preview status and per-model parameter differences require developers to verify support across models.

Mistral opens Responses API in public preview, borrowing OpenAI’s format for agent builders Photo: Merlin Lightpainting / Pexels The stateless endpoint supports multimodal inputs, JSON outputs, streaming and function calling on Mistral's hosted chat models Mistral AI https://cryptobriefing.com/markets/mistral/ wants switching AI providers to feel less like moving house and more like changing a lightbulb. The company has released its Responses API in public preview. The new interface lets developers build agents and applications on top of Mistral’s hosted chat models. The API is built to mirror OpenAI https://cryptobriefing.com/markets/openai/ ’s Responses format, so developers already working in that format won’t need to learn a new one from scratch. What the Responses API actually does The service launched in October 2026. It lives at the /v1/responses endpoint on Mistral’s Studio/API platform. Developers can work with it through official SDKs for Python and TypeScript. The API is stateless. Mistral’s servers don’t hold on to the conversation. Your application has to store the chat history and send it along with every request. On the input side, the API is multimodal. It accepts text, files and images, with images supplied either as URLs or as base64 data. Outputs can be structured as JSON. Instead of a free-form paragraph that a program has to parse, the model returns data in a predictable, machine-readable shape. The API also supports streaming. Responses arrive piece by piece as the model generates them, rather than in one block at the end. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Function calling is included too, and it is client-executed. The model can request that a tool be used, but your own code runs the tool and hands the result back. Reasoning parameters are available as well, though they vary depending on which model is in use. Mistral’s flagship mistral-large-4 plays a central role. It serves as the primary model in the API’s examples. How Mistral got here Mistral AI was founded in 2023 by former researchers from DeepMind and Meta https://cryptobriefing.com/markets/meta/ . Its pitch has leaned on features that set it apart from larger rivals, including stronger support for data residency and a commitment to open-weight models. The Responses API isn’t Mistral’s first step into agents. The company has shipped earlier agent-related releases that laid the groundwork for this one. This launch fits into a wider strategy to expand its developer tooling. The two focus areas are agentic features and compatibility with OpenAI interfaces. What this means for developers and the AI market The most immediate effect is lower switching costs. Because the API mirrors OpenAI’s Responses format, developers can more easily port existing workloads to Mistral. It also opens the door to hybrid setups. A team could, in principle, route some tasks to one provider and some to another without maintaining two entirely different codebases. Still, there are caveats worth flagging. The API is in public preview, so developers should treat it as a product that is still being shaped rather than a finished one. The stateless design gives teams control, but it also means more engineering work to manage conversation history on their own side. The model-dependent reasoning parameters deserve attention too. Developers building across several Mistral models will need to check which settings each one actually supports. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .