European LLM lab with open-weight models, a chat app and a low-cost API
Mistral is the lab to look at when the token bill matters more than the last point of benchmark score. Mistral Large is priced at $0.5 per million input tokens and $1.5 per million output, a fraction of the frontier models, and several of its models are open weights under Apache 2.0, so a fintech that cannot send data abroad can run them on its own GPUs. Le Chat is a credible ChatGPT alternative with a student price. For the hardest reasoning tasks, Claude and GPT still lead.
Are we wrong about your product? Tell us.
Mistral AI is a Paris-based model lab. It sells three things: an API for its models (Mistral Large, Medium, Small, the Ministral small models, Codestral for code, Voxtral for speech, OCR), the Le Chat assistant for consumers and teams, and open-weight models you can download and run yourself. Several current models carry the Apache 2.0 licence; others use a modified MIT or research licence, so check the model card before self-hosting commercially.
For an Indian team the two things that matter are price and deployability. The API undercuts OpenAI and Anthropic on most tiers, batch processing halves the price again, and cached input tokens cut input cost by up to 90%. The open-weight models mean an on-premise deployment is possible for data that must not leave the country.
Quick facts: Paris, founded 2023 · API, Le Chat and open-weight models · Mistral Large $0.5 in / $1.5 out per 1M tokens · Batch processing 50% off · Student Pro plan $5.99.
Mistral Large and Medium for quality, Small and Ministral (3B, 8B, 14B) for cheap, fast tasks on edge or batch jobs.
A code-completion and generation model with fill-in-the-middle, used in editor plugins and the Vibe coding agent.
Mistral Small, Large 3, Ministral 3 and others are published under Apache 2.0; download from Hugging Face and run on your own hardware.
A dedicated OCR model priced per 1,000 pages for invoices, KYC documents and scanned contracts.
Transcription and text-to-speech priced per minute, including a realtime transcription model.
Web and mobile assistant with web search, image generation and coding sessions; Free, Pro, Team and Enterprise plans.
Two price lists: Le Chat plans for people, and per-token API prices for builders. The page quotes Mistral Large as its example; other models have their own rows in the API table on the same page.
| Plan | Price | What you get |
|---|---|---|
| Le Chat Free | $0 | Limited messages, web searches and coding sessions |
| Le Chat Pro | $14.99 per month | Full access to Vibe for coding, more image generation; verified students $5.99 per month |
| Le Chat Team | $24.99 per user / month | Shared workspace, up to 30 GB storage per user, domain verification, data export |
| API: Mistral Large | $0.5 per 1M input tokens, $1.5 per 1M output | Batch processing 50% off; cached input up to 90% off |
| API: OCR, speech, tools | Priced per 1,000 pages, per minute and per call on the page | Rates per model on the pricing page |
Source: mistral.ai/pricing, checked 22 Oct 2026. Prices in US dollars excluding taxes; Indian cards are charged in USD and GST may be due under reverse charge for businesses.
In the order a small team would do them.
Set a monthly spend limit in the console before the key reaches a CI pipeline.
Run your evaluation set on Small and Medium first; most extraction tasks never need Large.
Batch requests are half price; nightly summarisation and backfills belong there.
A fixed 5,000-token policy document in every call is where the 90% cached-input discount pays.
If the answer is yes, test the Apache 2.0 open-weight model on your own GPU before building on the API.
If Mistral AI is not the right fit, these are the pages to read next.
Anthropic's models for the hardest reasoning, coding and agent work, with India plan prices on our page.
Choose when: Quality on complex tasks matters more than the per-token priceIndian lab with models built for Hindi and other Indic languages, speech included.
Choose when: Your users write and speak in Indian languages