Responses API

View Docs

Mistral's Responses API is in public preview for building agents. Each response returns the model's answer, its reasoning and its tool calls as separate items, making it easier than Chat Completions. It works with all hosted chat models including Large 4, is stateless (app keeps history), and compatible with OpenAI Responses format. Also serving DeepSeek V4.1 Flash (public preview) and GLM 5.3 Flash (generally available).

Links

Built by

Mistral AI

Cosigned by

Mistral’s Responses API lets applications use hosted Mistral models through the OpenAI Responses format. Its stateless endpoint accepts text, images, and files; supports streaming, structured outputs, reasoning, and client-executed function calls; and provides a separate endpoint for counting input tokens. For multi-turn conversations, the application sends prior messages and response items again with each request. The API documentation includes Python and TypeScript SDK examples as well as direct HTTP requests.

Anand Butani: You're probably still parsing reasoning out of one blob of Chat Completions text, and one malformed tool call quietly poisons the whole retry. Separate items let you retry just that step.
Deep: separate items make your logs easy to replay step by step
fatooom: Answer, reasoning and tool calls as separate items is the right shape for agents. When an agent can spend money, I want every tool call to be its own logged step, not something buried inside one completion.
Ghost: answer, reasoning and tool calls as separate items is lovely, now i can see the exact step my agent panicked 😭
John Rood: the payoff nobody's pricing in: typed items make two runs diffable. when a build goes sideways you walk a good run and a bad one item by item and the first call that diverged shows up, instead of guessing from the final answer.
Maé Prina: Mistral sort sa Responses API pour construire des agents. Chaque réponse montre séparément ce que le modèle dit, comment il a réfléchi et les actions qu'il lance. Petit détail, gros gain pour ceux qui construisent. Bien joué @MistralAI
Rohan Bhanot: Getting tool calls back as their own items is huge for anyone logging agent runs. Fishing them out of one blob was always messy
Sarthak: @MistralAI put its Responses API in public preview. It returns answers, reasoning and tool calls as separate items. I like this because it should make it easier to see what an agent did and handle each step in an app.
Shaoze Chen: 🧩 Mistral just shipped a Responses API, and the quiet headline is the format. It's in public preview, made for building agents: each response returns the model's answer, its reasoning, and its tool calls as separate items. Per Mistral's docs, it uses the OpenAI Responses format. If your agent already speaks that format and keeps its own history and functions, trying Mistral should be much closer to a config change than a rewrite. It's also stateless: your app keeps the history and runs the functions. More plumbing on your side, but the conversation history stays in your hands. The agent API layer is converging. Good. Less lock-in, more competition on the models themselves.
Vishal Singh: Nice to see the Mistral Responses API is now in public preview 🥳 In simple terms, this means it now returns answers, reasoning, and tool calls as separate items. This is particularly useful as it'll make it much easier to see what an agent did and handle each step in an app seamlessly. Nice upgrade, team 👌
Wei佳: Returning reasoning as its own item is the part I would actually build on.

Responses API

View Docs

Links

Built by

Mistral AI

Cosigned by

Anand Butani: You're probably still parsing reasoning out of one blob of Chat Completions text, and one malformed tool call quietly poisons the whole retry. Separate items let you retry just that step.
Deep: separate items make your logs easy to replay step by step
fatooom: Answer, reasoning and tool calls as separate items is the right shape for agents. When an agent can spend money, I want every tool call to be its own logged step, not something buried inside one completion.
Ghost: answer, reasoning and tool calls as separate items is lovely, now i can see the exact step my agent panicked 😭
John Rood: the payoff nobody's pricing in: typed items make two runs diffable. when a build goes sideways you walk a good run and a bad one item by item and the first call that diverged shows up, instead of guessing from the final answer.
Maé Prina: Mistral sort sa Responses API pour construire des agents. Chaque réponse montre séparément ce que le modèle dit, comment il a réfléchi et les actions qu'il lance. Petit détail, gros gain pour ceux qui construisent. Bien joué @MistralAI
Rohan Bhanot: Getting tool calls back as their own items is huge for anyone logging agent runs. Fishing them out of one blob was always messy
Sarthak: @MistralAI put its Responses API in public preview. It returns answers, reasoning and tool calls as separate items. I like this because it should make it easier to see what an agent did and handle each step in an app.
Shaoze Chen: 🧩 Mistral just shipped a Responses API, and the quiet headline is the format. It's in public preview, made for building agents: each response returns the model's answer, its reasoning, and its tool calls as separate items. Per Mistral's docs, it uses the OpenAI Responses format. If your agent already speaks that format and keeps its own history and functions, trying Mistral should be much closer to a config change than a rewrite. It's also stateless: your app keeps the history and runs the functions. More plumbing on your side, but the conversation history stays in your hands. The agent API layer is converging. Good. Less lock-in, more competition on the models themselves.
Vishal Singh: Nice to see the Mistral Responses API is now in public preview 🥳 In simple terms, this means it now returns answers, reasoning, and tool calls as separate items. This is particularly useful as it'll make it much easier to see what an agent did and handle each step in an app seamlessly. Nice upgrade, team 👌
Wei佳: Returning reasoning as its own item is the part I would actually build on.

Product name

Story

Product name

Story

Product name

Story

Product name

Story

Responses API

Product name

Story

Responses API