Back to Gap Map Model components / Inference code

Mistral Studio

closed / Overall score: 4.0(strong)

Mistral Studio, formerly La Plateforme, is Mistral AI's platform for running its models through an API. The API reference covers chat completion, fill-in-the-middle code completion, embeddings, classifiers, OCR, speech transcription and synthesis, and batch jobs, alongside endpoints for agents, workflows and observability. Beyond the hosted API, Mistral sells hybrid, dedicated and self-hosted deployments of the platform.

The ratings cover the model API. Studio's agent-building, workflow and evaluation tools sit on top of it and are not assessed here.

Openness

1 high confidence

1.0

engine: proprietary
source: closed
access: managed-cloud-API(hybrid, dedicated and self-hosted deployments sold commercially)
license: Proprietary

Mistral sells Studio as a hosted API and as hybrid, dedicated or self-hosted deployments, and none of them comes with published source for the platform. Its commercial terms grant only a limited, non-sublicensable license to use the products and forbid any attempt to discover their source code.

https://legal.mistral.ai/terms/commercial-terms-of-service recorded 2026-10-13
The commercial terms grant "a limited, non-exclusive, non-transferrable ... non-sublicensable (except to its End Users) license to access and use the Mistral AI Products" and bar customers from attempting "to reverse engineer, decompile, or otherwise attempt to discover the source code or underlying components (e.g., algorithms, weights, or systems) of the Mistral AI Products".
https://mistral.ai/products/studio recorded 2026-10-13
The Studio page, which mistral.ai/products/la-plateforme now redirects to, offers "Hybrid", "Dedicated" and "Self-hosted" deployment ("Retain full control over infrastructure and data residency") and a "Developer toolkit" of "APIs, SDKs, code samples"; its only GitHub link is mistralai/cookbook, and no source for the platform itself is linked.

Adoption

4 medium confidence

4.0

OpenRouter's provider page charts heavy daily traffic routed to Mistral's API, and Mistral names enterprise customers such as HSBC, ASML, CMA CGM and Stellantis. Neither is a count of the API's own users, and much of the OpenRouter traffic is for GLM models that Mistral also serves.

https://mistral.ai/customers recorded 2026-10-13
The customer stories page ("Bringing customers to the frontier") lists stories for HSBC, ASML, CMA CGM, AXA, Airbus, BMW, Cisco, Ericsson, the European Patent Office, Stellantis and TotalEnergies, among others; none gives a usage figure, and the stories cover Mistral's products generally rather than the API alone.
https://openrouter.ai/provider/mistral recorded 2026-10-13
OpenRouter's Mistral provider page states "Access 24 models served by Mistral through the OpenRouter unified API" and charts "Tokens processed on OpenRouter"; the embedded daily series for 2026-10-12 sums to about 119B tokens, led by mistralai/mistral-large-4-0-20261006 at about 58.4B and z-ai/glm-5.3-20260816 at about 50.7B.

Capability

4 medium confidence

4.0

Mistral's API covers more kinds of work than most hosted inference services, from chat and code completion to OCR, speech and batch jobs, and it is the first-party source for Mistral's own models. Its fastest models run slower than the fastest Fireworks serves, so it sits alongside Fireworks rather than above it.

https://artificialanalysis.ai/providers/mistral recorded 2026-10-13
The provider page says "Mistral offers 8 models" and "For output speed, the fastest models are GLM-5.2 (max) (223 t/s), Ministral 3 3B (215 t/s), and Mistral Medium 3.5 (163 t/s)."
https://docs.mistral.ai/api recorded 2026-10-13
The API reference navigation lists "Chat Completion", "Fim", "Embeddings", "Classifiers", "Files", "Models", "Batch", "Ocr", "Audio Transcriptions", "Audio Speech", "Realtime Sessions" and "Workflows" endpoints, with beta Agents, Conversations, Libraries, Connectors and Observability groups; the chat endpoint is POST /v1/chat/completions.

Verified 2026-10-13