Mistral API pricing: what every model costs per million tokens

Mistral API pricing is charged per million tokens, with input and output counted separately, plus per-page, per-minute and per-character prices for its OCR, transcription and voice models. On October 28, 2026 Mistral's pricing documentation lists Mistral Large 4 at $1.36 per million input tokens and $4.18 per million output tokens, while the smallest Ministral model costs $0.10 for both. This guide lists the main current prices, the discounts for batch processing and cached input, a worked monthly cost example, and the small print about preview models and aliases that can change what you pay.

Mistral API prices at a glance

Table: Mistral API prices per million tokens (Mistral's pricing documentation, standard tier, checked October 28, 2026)

Table: Mistral's specialized models (prices as marked)

Mistral Large 4: price and Public Preview

Mistral Large 4 is the newest flagship: a "state-of-the-art, open-weight, general-purpose multimodal model" with a mixture-of-experts design of 52 billion active parameters out of 1.05 trillion, a 1.6 billion parameter vision encoder and a 1 million token context window. Its model page dates it October 6, 2026 and marks it as Public Preview.

Two details matter for budgeting. First, the sale price of $0.68 and $2.09 that the model page and the pricing table showed on October 14, 2026, with no end date, no longer appears: on October 28 both list only $1.36 and $4.18. Second, Mistral's lifecycle policy says Public Preview models are "priced at the same rate as General Availability models" but allow silent updates and "may be retired before reaching" general availability, with one month's notice of deprecation instead of six.

Even so, Large 4 costs less per token than Mistral Medium 3.5 ($1.50 and $7.50), which is the most expensive text model on the table.

Batch, caching and other discounts

• Batch processing: Mistral's pricing FAQ says batch "reduces the price by 50%" for high-volume work that does not need an answer straight away. Its documentation says batch requests are queued and processed over a 24-hour period.
• Cached input: repeated prompts cost less. The FAQ says cached input "reduces input cost by up to 90%", and the table shows cached input at exactly a tenth of the input price for every text model. Our guide to prompt caching explains when a cache hit happens.
• Labs models: models with a labs- prefix are "free of charge" but experimental, can change silently, do not allow you to opt out of data collection and are not recommended for production.
• Priority Tier: an enterprise option with priority queueing and a 99.5% uptime SLA, which needs an agreement and custom rate limits set up by Mistral.

What a month costs: a worked example

Take a support chatbot handling 10,000 conversations a month, each with about 2,000 input tokens and 300 output tokens: 20 million input and 3 million output tokens.

Table: monthly cost for 20M input and 3M output tokens (standard prices, no caching or batch)

For comparison with other providers on the same token counts: OpenAI's GPT-6 Luna ($0.10 input, $0.50 output) comes to $3.50, and DeepSeek V4.1 Flash at peak hours ($0.30 and $1.20) to $9.60. Different companies' tokenizers split the same text into different numbers of tokens, so treat cross-provider comparisons as approximate; our guide to what a token is shows why. Our LLM API pricing overview and the guides to OpenAI API pricing and DeepSeek API pricing cover those providers in detail.

Aliases, open weights and what to check

Pin the version. Mistral's model names have aliases such as mistral-medium-latest, and its documentation warns that "using an alias may expose you to silent updates in model behavior and pricing". Use a fixed identifier like mistral-medium-3-5 in production.

Watch deprecations. General availability models get six months' notice before retirement; after that, requests return a 404 error.

Consider running it yourself. Several of these models are open-weight: Mistral's documentation lists Large 3, Small 4 and the Ministral 3 family under Apache 2.0. For small models, running them on your own hardware removes the per-token bill entirely; our guide to running an LLM locally shows how. For audio, Voxtral Mini Transcribe 2 at $0.003 a minute is cheaper than OpenAI's gpt-transcribe, and you can also transcribe audio locally for free.

For the consumer side, Mistral's chat assistant is now called Vibe (formerly Le Chat), with a Free plan and paid plans from $14.99 a month; those plans are separate from API billing.