What Is Mistral AI Studio (La Plateforme)? Free API Key, Pricing, Limits, Models, Turkish and Data
In Mistral AI Studio (console.mistral.ai) you get an API key and call Mistral models from code; the service people search for as La Plateforme is the subject of this page. New accounts start in Free mode: you get an API key without a credit card, but usage and speed are limited. For more, you pay as you go with prepaid credits. Pricing varies by model; for Mistral Small 4 it's $0.15 input and $0.6 output per million tokens. Turkish is among the languages where the language models are expected to perform strongly. You can turn off the use of your Studio and API data for model improvement from the Admin panel; the documentation doesn't state the default. The chat app (formerly Le Chat) is now a separate product called Vibe.
- What it does
- Developer console and API where you call Mistral models
- Free
- Free mode: no card, limited usage and speed
- Turkish
- Expected to be strong in the language models
- Training
- Turned off for the API in Admin; default not clear
- Zero retention
- On paid plans, for stateless endpoints
- Price
- Small 4: $0.15 input / $0.6 output (1M tokens)
What is Mistral AI Studio (La Plateforme) and what is it for? Can I use it without code?
!Name confusion: Mistral's current documentation calls the developer console and API “Studio” (console.mistral.ai); the name “La Plateforme” survives only in some URL paths (e.g. admin.mistral.ai/plateforme/limits). The documentation doesn't explicitly say the two names are the same product; this page covers that developer console and its API. The chat app Le Chat is now a separate product called Vibe and a separate page topic.
Mistral AI has three products: Vibe (an agent for everyday work and coding), Studio (API and Playground) and Admin (organization, billing, access). In Studio you generate API keys, try models in the Playground, build agents and workflows, and monitor usage. The API is how a program sends requests to Mistral: you send text and an API key (a secret password identifying your account), and the model returns a response. You can call text, audio and document (OCR) models with a single SDK.
Partly. The Playground in Studio lets you pick a model and send requests without writing code, and change settings like temperature and max tokens. Connecting it to your app requires the API and code. For everyday no-code chat, look at Vibe, which is a separate product.
Who it's for: developers who want to connect Mistral's open-weight models to their own apps; teams that want inference to run in the EU or US region (a regional inference option exists).
How do you get a Mistral API key?
- Create a Mistral account or sign in at console.mistral.ai (Studio).
- For new accounts, Studio comes activated in Free mode without asking for a credit card.
- Go to the API Keys page in the left menu and press Create new key.
- Give the key a name, choose an expiration date and set the connector access scope: only connectors shared in the workspace, or private and shared connectors.
- Copy the key right away and put it in a password manager: the full key is shown only once and can't be retrieved later.
› export MISTRAL_API_KEY="your-key-here"› $env:MISTRAL_API_KEY="your-key-here"› pip install mistralai› npm install @mistralai/mistralaiThese documents don't say whether there's phone verification, ID, age or a restriction on signing up from Turkey; check what the sign-up screen asks for. For the person managing the account, the “Admin” panel has user, workspace and role management.
How do you send your first request? A ready example
The Python example from the official “first API request” page (model names change often; current names are in the “Which model?” section):
› import os
from mistralai.client import Mistral
client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])
response = client.chat.complete(
model="mistral-large-latest",
messages=[{"role": "user", "content": "What is Mistral AI?"}],
)
print(response.choices[0].message.content)› curl https://api.mistral.ai/v1/chat/completions \
-H "Authorization: Bearer $MISTRAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "mistral-small-latest", "messages": [{"role": "user", "content": "Hello, Mistral!"}]}'mistral-small-latest (fast) or mistral-large-latest (higher quality) from the Model list, write the message and press Send. Temperature, max tokens and Top P are set in the panel on the right.| Error | What to do, per the official docs |
|---|---|
| 401 Unauthorized | The key is wrong or not set. In the terminal you're working in, check that the variable is filled with echo $MISTRAL_API_KEY. |
| Monthly spending limit reached | If the organization's monthly spending limit is reached, API access may be suspended until the start of the month or until an admin raises the limit. Access can also be affected if recent invoice payments failed. |
| Rate limit error | Your own limits are on the Admin Panel › API › Limits page; for higher limits, write to support at help.mistral.ai saying which model, how many requests per second and how many tokens per month you'll use. |
Which Mistral model? The differences
| Model (API name) | Mistral's own description | Context | License |
|---|---|---|---|
Mistral Large 4 (mistral-large-4) | Open-weight, general-purpose multimodal (text + image) model; speed 5/5, performance 4/5 | 1M | Not specified |
Mistral Medium 3.5 (mistral-medium-3-5) | Top-tier multimodal model for agent and coding work; performance 4.5/5, speed 2/5 | 256k | Modified MIT |
Mistral Small 4 (mistral-small-2603) | Hybrid model combining instruction following, reasoning and coding in one efficient model | 256k | Apache 2.0 |
| Ministral 3 (14B, 8B, 3B) | Small, efficient text + image models; 3B is the “tiny” one | 256k | Apache 2.0 |
Codestral (codestral-2508) | Code completion model (late July 2025) | 128k | Not specified |
OCR 4.1 (mistral-ocr-4-1) | Document OCR; paragraph-level bounding boxes and confidence scores. $4 per 1,000 pages ($5 with annotations) | None | Not specified |
Voxtral Mini Transcribe 2 (voxtral-mini-2602) | Audio-input model for transcription only. $0.003 per minute | None | Not specified |
Voxtral TTS (voxtral-mini-tts-2603) | Text to speech; zero-shot voice cloning; 9 languages. Output $16 per million characters | None | CC BY-NC 4.0 (non-commercial) |
Mistral Moderation 2 (mistral-moderation-2603) | Content moderation, 128k context, jailbreak detection. Free | 128k | Not specified |
- Fast, cheap text work (labeling, short summaries) → Ministral 3 3B or 8B, Mistral Small 4.
- Best quality, long context → Mistral Large 4 (1M) or Medium 3.5 (agents and coding).
- Coding help → Codestral. Reading scanned documents → OCR 4.1. Transcribing audio files → Voxtral Mini Transcribe 2.
- Third-party models are hosted too (e.g. Z.ai GLM 5.3, 1M context); their prices are on the same page.
!Models prefixed with “Labs” are experimental and available for a limited time; they're open only to commercial customers and excluded from zero data retention. Under the terms, inputs and outputs of Labs and preview models may be used for training. Before using a Labs model in production, see the “My data” section.
The model list, prices and licenses change often; this section was checked on October 15, 2026. Speed and performance scores are Mistral's own 0–5 ratings from its model comparison page.
Is the Mistral API free? Free mode, limits and pricing
- Default for new accounts
- No credit card needed
- API key, Playground, included monthly usage
- Limited usage and speed (numbers aren't in the docs, they're in the console)
- No zero data retention (paid plans only)
- Beyond the included monthly usage
- Prepaid credits: buy them or use a gift code
- You can set a monthly spending limit
- Not a separate API plan; added to your current plan
- The same plan applies to Vibe, Studio and the API
- Included monthly usage is shared across Studio, API and Vibe Code
- Zero data retention (on request)
| Model (API name) | Input (1M tokens) | Output (1M tokens) |
|---|---|---|
Mistral Large 4 (mistral-large-4) | $0.68 (sale; regular $1.36) | $2.09 (sale; regular $4.18) |
Mistral Medium 3.5 (mistral-medium-3-5) | $1.5 | $7.5 |
Mistral Small 4 (mistral-small-2603) | $0.15 | $0.6 |
Ministral 3 14B (ministral-14b-2512) | $0.2 | $0.2 |
Ministral 3 8B (ministral-8b-2512) | $0.15 | $0.15 |
Ministral 3 3B (ministral-3b-2512) | $0.1 | $0.1 |
Codestral (codestral-2508) | $0.3 | $0.9 |
OCR and audio models are priced per page, minute or character rather than per token; see the “Which model?” section. The price shown for Mistral Large 4 is a “sale price”; the page doesn't say how long it lasts.
Free mode's numeric limits (tokens per minute, requests per second) aren't in the documentation; they appear per model on the Admin Panel › API › Limits page. A monthly spending limit can be set at the organization and workspace level; if the limit is reached, API access may be suspended.
Prices are in dollars; on the pricing page you can choose US dollars or euros as the currency. Local-currency pricing and tax treatment aren't stated; for student and teacher discounts see Mistral's help article; it doesn't say whether they apply to the API.
Does Mistral work in Turkish?
| Text replies (language models)Yes | Mistral's supported languages page lists Turkish in the “Middle East” group, among languages with strong expected performance. There's no per-model Turkish benchmark; the list is based on the models' general training. |
|---|---|
| OCR (document reading)Yes | In the OCR language list, Turkish is in the Central Asia group. |
| Speech → text (Voxtral)Unclear | The supported languages page covers only language models and OCR; no list including Turkish appears for the audio model. |
| Text → speech (Voxtral TTS)Unclear | The model page says “9 languages”; it doesn't list which ones. |
| Turkish Studio interfaceUnclear | Not stated in the official documentation; if there's a language setting on screen, check there. |
Does Mistral use my data for training? Zero data retention
Mistral's help article: in some cases inputs and outputs may be included in model training programs. For Studio and the API, you turn off the “Anonymous improvement data” toggle in the Privacy menu of the Admin panel. The Vibe and API toggles are separate; turning off one doesn't turn off the other. These documents don't say whether the API default is on or off.
Mistral doesn't use customer data and outputs to train AI models, except when: training is off by default in a product and you turn it on, or it's on by default and you don't turn it off; you give feedback; the contract says otherwise; you use Labs or preview models.
Enabled on request on paid plans, for stateless API endpoints (such as chat completions, embeddings, OCR, audio, moderation). Agents, Batch files, Conversations, Libraries, Files API, Vibe Work and Chat are out of scope; Labs models are excluded too.
Regional inference (EU or US) is optional and costs 1.1 times the list price. If you don't specify a region, the global endpoint is used; the control plane, such as account, keys and billing, isn't regional.
The official documentation shows no Turkey-specific KVKK (Turkey's Personal Data Protection Law) text and no account deletion steps. This is not legal advice; if you'll process personal data, read the terms and the DPA yourself.
Frequently asked questions
How do I get a Mistral API key, and is it free?
You create an account at console.mistral.ai (Studio) and generate a key on the API Keys page. Free mode doesn't require a credit card, but usage and speed are limited.
What are the Mistral API free limits?
The official documentation says Free mode has usage and speed limits but gives no numbers. Your tokens-per-minute and requests-per-second limits per model appear on the Admin Panel, API, Limits page.
Are Mistral La Plateforme and AI Studio the same thing?
Current Mistral documentation calls the developer console and API Studio; the name La Plateforme survives only in some URL paths. The documentation doesn't explicitly say the two names are the same.
How much does the Mistral API cost?
Per token: for Mistral Small 4, $0.15 input and $0.6 output per million tokens; for Medium 3.5, $1.5 and $7.5; for Ministral 3B, $0.1. OCR is $4 per 1,000 pages. There's no local-currency pricing.
Does Mistral speak Turkish?
Mistral's supported languages page lists Turkish among the languages with strong expected performance for the language models; it's also on the OCR list. These documents don't cover Turkish for the audio models.
Does the Mistral API use my data for training?
The commercial terms say it won't, except with Labs and preview models and when you turn training on. For Studio and the API, you turn off the training toggle under Privacy in the Admin panel; the documentation doesn't state the default.
Related guides
Sources
- Platform overview
- Studio and API key
- First API request
- API pricing
- Model selection guide
- Usage and limits
- Subscriptions
- Supported languages
- Zero data retention
- Regional inference
- Opting out of training (help article)
- Commercial Terms of Service
Not clear in the official sources: The name relationship (La Plateforme and Studio) · Free mode's numeric limits · the training default for the API · phone, ID and country requirements at sign-up · Turkish in the audio models · interface language · account deletion · local-currency pricing and tax · how long the Mistral Large 4 sale lasts · whether the student discount applies to the API. Each is covered in the relevant section along with what you can do.