Mistral AIBetaVerified 2026-10-30

Mistral Responses API

New (public preview): Mistral models behind the OpenAI Responses format at POST /v1/responses. Output arrives as typed items (messages, reasoning, function calls), with JSON Schema output and streaming. Stateless: you resend history each turn.1

CompareDocs

PricingUSD, as published

ItemValue
List pricePer-model token price2
Free tierUnverified

No separate Responses price: you pay the model’s API token price. Example: mistral-large-4 $0.70 per 1M input, $0.07 cached input, $2.10 output.2 [Verified 2026-10-30 on https://docs.mistral.ai/inference/pricing]

Not in the cost calculator: Billed at the chosen Mistral model’s API token price (e.g. mistral-large-4 $0.70 in / $2.10 out per 1M); the calculator does not list Mistral models yet.

LimitsPer request

LimitValue
Context windowNot published
Rate limitPer org & model (console)

Stateless: no stored conversations, so your app keeps history and resends earlier output items. Only client-executed functions are listed (no hosted tools).1 [Verified 2026-10-30 on https://docs.mistral.ai/studio/conversations/responses]

Rate limits are set per Organization and per model (tokens per minute and per month), shown in the Admin Panel under API › Limits. Mistral publishes no fixed public numbers and no separate /v1/responses limit.4 [Verified 2026-10-30 on https://docs.mistral.ai/admin/billing-usage/usage-limits]

StatusBeta

  • Public preview. Mistral asks for issues and feedback through support.1 [Verified 2026-10-30 on https://docs.mistral.ai/studio/conversations/responses]

Auth

MethodAPI key (Bearer)
DetailsMistral API key as Authorization: Bearer $MISTRAL_API_KEY on POST https://api.mistral.ai/v1/responses.3 [Verified 2026-10-30 on https://docs.mistral.ai/api/endpoint/responses]

FeaturesFrom lib/apiFeatures.js

YesNoN/ANot published
CapabilityStateNote
Structured JSON outputJSON Schema via text.formatYesJSON Schema via text.format
Tool / function callingClient-executed functionsYesClient-executed functions
StreamingResponses eventsYesResponses events
Official SDKsmistralai (Python), @mistralai/mistralaiYesmistralai (Python), @mistralai/mistralai
Free tierUnverified
BatchUnverified
MCP supportUnverified
Fine-tuningUnverified
Open source / self-hostUnverified
SLA publishedUnverified
EU / data residencyUnverified

Fit and safety

Good for

  • Good for moving an existing OpenAI Responses integration to Mistral models.

Watch out

  • Preview: field shapes may change, and features beyond the supported list (stored state, hosted tools) are not available.

How it works

  1. Text input and output (strings and message arrays); image input (URLs and base64) with a vision model; file input (HTTPS URL, base64 or uploaded file ID).
  2. Structured output: JSON mode and JSON Schema via text.format.
  3. Streaming Responses events for text and function arguments; handle response.completed, response.failed, response.incomplete and error.
  4. Function calling: client-executed functions, including namespaces. Reasoning: reasoning.effort and plaintext reasoning, depending on the model.
  5. POST /v1/responses/input_tokens counts input tokens before you generate.
  6. SDKs: pip install -U mistralai, npm install @mistralai/mistralai@latest (client.responses.create).

Compare

Put Mistral Responses API side by side with other APIs on price, limits and features. Open the comparison table →

Deciding between a decision model and a generative LLM? See Decision model vs LLM: when to use which.

SourcesFootnotes

  1. 1Mistral docs: Responses APIdocs.mistral.ai/studio/conversations/responsesVerified 2026-10-30
  2. 2Mistral API pricingdocs.mistral.ai/inference/pricingVerified 2026-10-30
  3. 3Mistral API reference: Responsesdocs.mistral.ai/api/endpoint/responsesVerified 2026-10-30
  4. 4Mistral docs: Usage and limitsdocs.mistral.ai/admin/billing-usage/usage-limitsVerified 2026-10-30
  5. 5Mistral launch post on Xx.com/MistralAI/status/2115443720371814478Cited