Skip to content
Neural Minute

Review · AI chatbots and companions

Mistral AI Review 2026: Open Weights at $14.99 a Month

By Maxime Aulne, Founder and Editor

Mistral AI is free to start, $14.99 a month for Pro and $24.99 a seat for teams. What Vibe opens, what the open weights allow, who should pay.

Minute Verdict

Mistral AI sells three things on one account: Vibe, the assistant that used to be called Le Chat, an API billed per million tokens, and open weights a company can download and run on its own machines. Vibe is free with caps, Pro costs $14.99 a month, Team is $24.99 per user, and the flagship model bills $0.70 per million input tokens against $2.10 per million output tokens.

Buy Pro if you want research, documents and coding in one place, and take the API instead if the work is programmatic. Skip it if you need video generation, a message count before you pay, or a discount for paying a year up front.

Best for: a European team that needs the model inside its own infrastructure, and a developer who wants weights to fine-tune.

Not for: a buyer who wants published quotas and an annual price, or a creator who needs video.

From: $14.99/mo billed monthly, checked November 17, 2026

What Mistral AI does

Mistral AI opened in Paris in 2023 with a paper argument: open weights beat black boxes for most uses, because a company can audit, adapt and host the model itself. Three years later the catalogue follows that line. Large 4, the flagship, ships under the Mistral Ecosystem Open Weight License, a modified MIT licence that allows modification, fine-tuning and commercial use, with one condition: a business that resells the model as a service has to buy a commercial licence first. Small 4 and the three Ministral 3 sizes ship under Apache 2.0.

The consumer product is Vibe, renamed from Le Chat on May 28, 2026. Work Mode plans a task, takes sign-off, then works across connectors such as Google Workspace, Outlook, SharePoint, Slack and GitHub, and it can schedule a prompt to run daily or weekly. Code Mode runs coding sessions in isolated sandboxes from the web app, and the same agent ships as a VS Code extension and a terminal CLI. Chat mode, the old Chat window, is being sunset, with history carried into Work Mode.

Under the interface, the lineup splits by job. Large 4 is a mixture of experts at 1.05 trillion parameters total and 52 billion active, fluent in more than 160 languages, and trained from scratch on European infrastructure. Medium 3.5 is a 128 billion dense model that absorbed the Magistral reasoning and Devstral coding lines, and it is already listed for retirement on April 30, 2027. Small 4 folds instruct, reasoning and coding into one model at 119 billion parameters with 6.5 billion active. OCR 4.1, Voxtral for speech and Codestral for code complete the list, each billed on its own meter.

The pitch to a company is sovereignty rather than features. Mistral sells private deployment and dedicated compute alongside the API, which is the part of the offer a US lab cannot match inside the European Union.

On this page

What you get for the money

Plan Price What it includes The limit that matters
Free $0 Vibe on web and mobile, limited messages and web searches, limited coding sessions, basic image editing, 100+ connectors No count published, only the word limited
Pro $14.99/mo, taxes excluded More messages and searches, all-day coding in the CLI, the IDE and the web, spreadsheets, slides, better image generation, chat and email support Fair usage limits, and no annual price
Education $5.99/mo Pro for verified students at accredited institutions, up to twelve months Only for accounts that never used Vibe or Le Chat
Team $24.99/user/mo, taxes excluded Pro plus a shared workspace, admin controls, up to 30GB of storage per user, 200 Flash answers a day Priced per seat, no annual view
Enterprise Quoted Private deployment, custom models and agents, audit logs, SAML SSO, white label No public price

The grid opens on the tax-excluded view, with a second view that adds local tax and a switch between dollars and euros. There is no annual toggle. Twelve months of Pro cost $179.88 and two Team seats $49.98 a month, with nothing to gain from paying up front.

The API is where the numbers get sharper. Large 4 bills $0.70 per million input tokens, $0.07 when the input is cached, and $2.10 per million output tokens. Small 4 bills $0.15 and $0.60, the 14B Ministral bills $0.20 on both sides, and Codestral bills $0.30 in and $0.90 out. OCR 4.1 charges $4 per 1,000 pages and Voxtral charges $0.0035 per minute of transcription. Batch work halves every rate, and a month of Pro buys roughly seven million output tokens of Large 4: the plan is for a person typing, the meter is for a program running.

Where it falls short

The caps are the first problem. Free and Pro are described with multipliers, six times the free allowance for messages and searches, forty times for image generation, and fair usage limits apply on top. Nobody can work out in advance how many prompts $14.99 buys, which makes the plan hard to budget.

There is no annual price anywhere on the grid. Every competitor in this category sells a year up front at a discount, and Mistral sells month to month only, so the twelve month bill is $179.88 with no lever to pull.

The model calendar moves fast. Medium 3.5 is already scheduled for retirement on April 30, 2027, behind Large 3, Medium 3.1, the Devstral 2 family and the earlier Magistral sizes. Anyone who pins a model name in production needs to read that lifecycle table before shipping.

The licence on the flagship is open, not permissive. Large 4 allows commercial use, modification and fine-tuning, but a business that wants to resell it as a model service needs a separate agreement with Mistral. What a plan does not include: video generation, which is absent from the product, and OCR or speech, billed on a separate meter. Image generation is capped per plan, so a design team will hit the ceiling rather than pay for more.

Mistral AI alternatives

  • ChatGPT, $8 a month on Go and $20 on Plus, the widest app and connector surface in this group, with image generation on the free tier: our ChatGPT review.
  • Claude, $20 a month on Pro or $17 billed yearly, the pick when long documents and code review matter more than the size of the model catalogue: our Claude review.
  • Google Gemini, $4.99 a month on Google AI Plus, the cheapest way into a frontier assistant already wired into Workspace: our Google Gemini review.

Who it is for: Mistral AI

Mistral AI suits a buyer who cares where the model runs. A European company with a residency requirement gets weights it can host, a European API and a private deployment offer, which is the one thing the US labs cannot sell inside the bloc. A developer gets a multimodal flagship at $0.70 per million input tokens with open weights behind it, so the same model can be called today and self-hosted later.

It does not suit a creator. There is no video generation, image generation is capped, and the creative work here is document and speech processing rather than pictures. It does not suit a buyer who wants a quota written down either, or a company that wants a year of service at a discount.

If the question is a general assistant rather than sovereignty, the AI chatbot reviews we track put the same features side by side, and the choice comes down to the connectors a team already uses.

How we checked

This is a research review. We have not run Vibe, so there is no measured task, no timing and no judgement on output quality beyond what Mistral publishes. Every amount was read on November 17, 2026, on mistral.ai/pricing in both tax views and both currencies, and on the API pricing page in the documentation. Model sizes, licences and retirement dates come from the models overview read the same day, the rename and the plan prices from the announcement of May 28, 2026, and the Large 4 licence from its open-weight release post. Competitor prices come from their own pages, read between November 12 and November 15, 2026. Our how we test page covers what a hands-on pass would add.

What to do next

Start on Free and put a real week of work through Vibe, including one connector and one scheduled task, because the connectors are what make the agent useful and the caps are what decide the plan. Move to Pro only when the caps bite. If the work is code that runs without you, price it on the API first: at $2.10 per million output tokens a batch job can cost less than the subscription it replaces.

Before paying, read the licence of the model you plan to depend on, and check its retirement date. For Large 4 this matters twice: the weights are open, but a service built on top of them needs a commercial agreement.

FAQ

Is Mistral AI free? Yes. The Free plan costs $0 and carries Vibe on web and mobile, limited messages and web searches, limited coding sessions and more than 100 connectors. The API has no free tier in the grid, though Mistral keeps a free Experiment plan in Studio for testing.

How much does Mistral AI cost? Pro is $14.99 a month excluding taxes, and verified students pay $5.99 for up to twelve months. Team is $24.99 per user a month, and Enterprise is quoted. No annual price is published. On the API, Large 4 bills $0.70 per million input tokens and $2.10 per million output tokens.

Is Mistral AI open source? The weights are open, not all of the licences. Small 4 and the three Ministral 3 sizes use Apache 2.0. Large 4 uses the Mistral Ecosystem Open Weight License, a modified MIT licence that allows modification, fine-tuning and commercial use, but a business that resells the model as a service needs a commercial licence first.

What happened to Le Chat? Le Chat was renamed Vibe on May 28, 2026. Plans, history and settings carried over, and the old chat window became Work Mode inside an agent that also covers coding, with a web app at chat.mistral.ai, a VS Code extension and a terminal CLI.

Which model should I pick? Medium for most tasks and coding, Small when the bill matters, Large 4 when the task needs the largest open-weight model, OCR for documents, Voxtral for speech and Codestral for code. Watch Medium 3.5 in particular: it is already on the retirement table for April 30, 2027.

Can I run Mistral models on my own servers? Yes, for the open-weight families. Download the weights from Hugging Face and serve them with vLLM or SGLang, or use Mistral’s private deployment offer. Large 4 is also sold as a hosted API, and the two routes are priced separately.

Sources

Sources

Update log

November 17, 2026: first publication, research review. The four plan cards read in both tax views and both currencies, the API rates for the nine model families read on the same day, and the licence and retirement facts taken from the models overview and the release posts.

Newsletter

One email when a review lands.

That is when the promo codes are still fresh, before the prices move.

Nothing else. Unsubscribe in one click.