Mistral Pricing (2026): API Costs and Vibe Seats

Mistral pricing for paid text generation starts at $0.1 per million input tokens and $0.1 per million output tokens for Ministral 3 3B; the worked app month below costs $13.50 on Mistral Small 4 or $150 on Mistral Medium 3.5. For people, Le Chat is now Vibe: Pro costs $14.99/month and Team costs $24.99/user/month before tax. Start a bounded app on Small, buy a seat for hands-on coding, and use regional inference or a private deployment when location requirements decide the purchase.

Prices verified against Mistral's pricing page, October 2026, with the live subscription page and API price sheet checked on 6 October 2026. Prices are in US dollars; calculated bills are illustrations using the stated token volumes, before tax and any applicable credits.

Mistral Pricing: API Rates at a Glance

Mistral's hosted API charges for what your app processes; a Vibe subscription buys access for a person. An API lets your software request model output. Tokens are the pieces of text it processes: input is what you send, output is what it generates, and cached input is previously processed material reused at the listed lower rate.

This table reproduces every model row in the current API price sheet, using Standard processing with regional inference switched off. Text and code rates are per million tokens. Other units are written in the cells; a dash means the page lists no separate output charge, not that every associated service is free.

Source: Mistral API pricing. Z.ai GLM 5.3 is a third-party model hosted by Mistral. The other labels are reproduced as the price sheet names them; Codestral is displayed as Codestral, rather than an older version name imported from another provider.

Large 4's sale needs a separate budget assumption. The page strikes through original rates of $1.36 input, $0.14 cached input and $4.18 output, replacing them with the sale rates above. It does not state an expiry date in the retrieved price sheet. Keep the original rates in your forecast for work that may continue beyond the promotion. The subscription page's generic “Mistral Large” example of $0.5 input and $1.5 output matches Large 3, so it should not be applied to Large 4. Source: version-specific price sheet.

Mistral API Pricing: Discounts and Extra Charges

Batch and caching lower different parts of the bill. Batch is asynchronous processing for work that can wait; Mistral advertises a 50% discount. Cached input can save up to 90% on repeated prompts. Neither reduces the fresh-output tokens your agent generates during an ordinary interactive request. Source: Mistral pricing terms.

Use Batch for offline classification or document queues where the result does not need to appear immediately. Use caching when repeated instructions or context qualify for reuse. An agent that writes long answers can still spend most of its money on output, even with a high input-cache hit rate.

Priority Tier is a paid service mode for eligible realtime traffic. Its 1.75x Standard multiplier is a 75% premium on input, output and cached tokens, applied after the cache discount. Access and limits depend on the supported model and your arrangement with Mistral. Source: Priority Tier pricing.

Regional inference adds 10% to Standard list pricing. Its feature restrictions matter as much as the surcharge: regional endpoints do not support Batch. The EU purchase decision below explains what the regional boundary covers. Source: regional inference.

Mistral OCR Pricing and Other Specialist Meters

Document and speech workloads need their own units. OCR, optical character recognition that turns document pages into usable content, is priced by pages. Transcription is priced by minutes; Voxtral TTS, text-to-speech generation, is priced by generated characters. Applying a million-token rate to those jobs produces the wrong estimate.

Codestral Embed creates representations of code for retrieval and similarity search. Its price sheet has input and cached-input rates without an output-token row. For a document assistant, count the document-processing step and the answering model separately whenever both are used. The free Moderation 2 row is a specific service, not a free allowance for every generative model. Source: model price sheet.

Mistral AI API Pricing: A Worked Monthly Bill

The requested month costs $13.50 on Small 4 or $150 on Medium 3.5. Assume 50 million fresh input tokens and 10 million output tokens, all Standard global requests, with no cached input, Batch discount, tool charges, credit offsets or tax. The calculation is input millions multiplied by the input rate, plus output millions multiplied by the output rate.

For Mistral Small 4, input costs 50 × $0.15 = $7.50. Output costs 10 × $0.6 = $6.00. Total: $13.50/month.

For Mistral Medium 3.5, input costs 50 × $1.5 = $75. Output costs 10 × $7.5 = $75. Total: $150/month.

These are calculations from Mistral's current rates, not measured app spending. Medium costs $136.50 more per month at these fixed volumes. Its name does not make it cheaper than every “Large” model: the exact version and price row decide.

For a solo builder extracting fields or classifying tickets, Small is the starting point if its answers pass the app's acceptance checks. For an agent doing coding or multi-step professional work, evaluate Medium when Small's corrections and retries erase the saving. A cheaper token rate only helps if the task gets finished correctly.

Mistral vs OpenAI, Anthropic and Google

Medium 3.5 costs less per token than the selected OpenAI and Anthropic models, while Google's selected model is cheaper during its promotion. The comparison uses models positioned for coding or agent work, rather than claiming identical capabilities. The monthly column holds the same fresh-input/output volumes constant; actual task consumption and acceptance can differ.

Rates below are per million tokens, Standard processing, verified October 2026. OpenAI uses its explicitly labeled short-context row; longer-context rates differ. Google includes thinking tokens, the tokens generated while reasoning, in its output price.

OpenAI GPT-6.1 Sol is the comparable Sol API option here. Its Standard long-context row is $4 input and $15 output, so the short-context calculation should not be reused for that traffic. Use the OpenAI API pricing guide to budget its other model and tool meters. Source: OpenAI pricing.

Anthropic Claude Sonnet 5.5 is listed for coding and agents at the rates in the table. A buyer already meeting acceptance checks with it should justify switching by total completed-task cost, including migration and corrections, rather than the token-price difference alone. Its caching and other model rates are covered in the Claude API pricing guide. Source: Anthropic pricing.

Google Gemini 3.8 Flash is the lowest listed price in this selected comparison through 31 December 2026. Google states that prices become $1.50 input and $7.50 output on 1 January 2027; the same example then costs $150, matching Medium's current calculated bill. A forecast extending into that period should use the scheduled rates. This is a pricing consequence, not a benchmark verdict. Source: Google Gemini pricing.

Mistral AI Pricing: Le Chat and Vibe Plans

Le Chat and Mistral Vibe are now one subscription decision. Vibe is Mistral's agent product for productivity and coding. Current documentation separates Work, the web/mobile experience with a fast/think toggle, from Code, development in a terminal, editor or remote session. Existing Le Chat accounts and plans carry over; you do not need to buy a second Le Chat plan alongside Vibe. Sources: current Vibe overview, Le Chat transition.

Free: Occasional Work and Evaluation

Free is enough when occasional chat and limited experimentation meet your needs. It includes web/mobile access, limited messages and searches, image generation, Studio model testing, 100+ connectors and $10/month in API credits. Document access is limited and scheduling allows up to five tasks. The page does not publish one universal daily message or search count. Source: Mistral plans.

Free coding access needs care. The detailed interface matrix does not include the CLI or IDE on Free, while broader wording mentions limited coding. Vibe Code Web documentation gives two sessions per day where access is enabled. Treat that as the documented web-session allowance, not a promise that every Free account has full terminal or editor access. Sources: plan matrix, Web session limits.

Pro: $14.99/Month for the Person Doing the Work

Pro is the solo coding seat, subject to fair usage. At $14.99/month before tax, the standard plan advertises $30/month in API credits, more complex tasks, coding in the CLI, IDE or web, and chat/email support. Source: Mistral Pro.

Its published limits are up to six times Free messages, five times web searches and forty times image generations, with 15GB document storage and up to 1,000 project folders. Scheduled tasks are described as unlimited subject to fair usage. The matrix includes Canvas, hooks that run commands around agent turns, remote coding, Teleport session handoff and session observability. Those features justify a seat when they fit the way you work; “all-day coding” is not an unlimited-token guarantee. Source: detailed plan comparison.

The advertised credits can reduce eligible API spending, but the public page does not explain every allocation condition or promise rollover. Keep the app's gross usage estimate separate, then apply the credit actually available in its billing balance. Do not treat the difference between the advertised credit and seat price as cash back.

Education: $5.99/Month With Eligibility Conditions

Verified students can pay $5.99/month for Pro's Education offer. Its feature list shows $15/month in API credits, rather than standard Pro's $30. The pricing-page eligibility note limits it to 12 months, students at accredited higher institutions and people who have not previously used Vibe or Le Chat; taxes are extra. Source: Education offer and eligibility.

Team: $24.99/User/Month for Organization Features

Team earns its premium through shared administration, not a universally larger message multiplier. It costs $24.99/user/month before tax, with 30GB storage per user, domain verification, knowledge connectors and data export. Its listed message/search/image multipliers match Pro. Source: Mistral Team.

Team also includes the paid coding and creation tools, up to 1,000 project folders, and scheduled tasks subject to the same unlimited-with-fair-usage wording. The pricing FAQ separately gives 200 Flash answers per day on Team versus 150 on Pro. That feature-specific allowance is a reason to upgrade if you need it; it does not establish a higher multiplier for every message. Source: plan matrix and usage FAQ.

The live calculator defaults to a minimum of two users and displays a rounded $50/month. The arithmetic is 2 × $24.99 = $49.98 before tax. That display is a seat estimate, not an additional fee or a $50 API-credit allowance. No separate Team credit amount is promised on the current plan card. Source: Team card and calculator.

For a company that codes, pay the $10/user/month increment over Pro when those organization features are required. Skip the upgrade if you only want a higher message multiplier. Team's matrix does not include SAML single sign-on or audit logs; those sit on Enterprise.

Enterprise: A Quote for Private Deployment and Controls

Enterprise is the private-deployment purchase. Mistral gives no public fixed price. It lists custom models, agents and workflows, audit logs, SAML single sign-on, white labeling, an Admin API, and deployment in your own cloud or on premises. The FAQ also lists custom service commitments and dedicated support. Source: Mistral Enterprise.

Choose it when the company needs that deployment/control boundary. A Team seat is not a substitute for private infrastructure, and a France-headquartered supplier alone does not establish where every workload runs.

Annual Billing, Overages and Refunds

Monthly billing is the safer starting point while you establish usage. Mistral's help center states a 20% annual discount for Pro and Team, while the current public price page displays monthly prices without an exact annual invoice quote. Confirm the actual annual charge at checkout. The discount implies a break-even of 9.6 monthly payments before tax and rounding, calculated as 12 × 0.8; expected continuous use for ten months or more can favor annual billing. Source: billing frequencies.

Pro and Team can extend usage beyond plan limits with pay-as-you-go credits at API rates. Set an overage budget separately from the seat fee. The pricing page does not state bundled-credit expiry or rollover terms, so unused monthly credits should not be assumed to accumulate. Source: usage extension and credits.

For web subscriptions, cancellation within 14 days of the initial purchase triggers an automatic refund under Mistral's published policy. After that window, cancellation leaves access until the billing period ends and does not trigger a refund; the window does not apply to renewals. Mobile-store and carrier purchases follow their purchase channel. Cancelling Vibe does not cancel the separate API plan. Sources: refund policy, subscription management.

Is Mistral Free? Limits That Matter

Free access is useful for learning and occasional work; it is not a production-throughput commitment. Someone who stays within the free product's chat, scheduling and enabled coding allowances, and whose eligible API experimentation fits the available credit, may never need a paid seat.

The API caps requests per second, tokens per minute and tokens per month at the organization level across workspaces, with limits that differ by model. The official help page directs you to your account's Admin Limits page for the exact figures. It does not provide a universal public free-token quota to copy into every app's forecast. Source: Mistral API rate limits.

Pay-as-you-go enables Tier 1. The published table lists Tier 2 above $20 billed, Tier 3 above $100, Tier 4 above $500, and support contact for higher limits after $2,000. These thresholds track cumulative billed consumption: buying prepaid credit does not itself raise throughput. A funded account can still hit a rate limit. Source: tier requirements.

Mistral AI Free Models: Separate Weights From Hosted Usage

A model's availability does not make its hosted API calls free. The current price sheet explicitly marks Mistral Moderation 2 as free, while it charges for the generative rows above. If you host model weights yourself, the model files and the compute bill are different purchases: evaluate the exact model license and your own infrastructure cost rather than assuming a zero-cost service.

For an app, use the API's posted rate and the credit balance actually assigned to the account. For occasional personal work, stay on Free until a named limit or paid feature blocks something you need. Source: hosted pricing.

A Coding Plan Key Can Stop Your App

Use an application API key for automation, not a Vibe plan key. Mistral's help center says a key created through Code > Vibe Code CLI can draw from the monthly Vibe budget and stop when that budget is exhausted. A Studio/pay-as-you-go key is governed by API rate limits instead. That distinction explains why a coding-plan quota can interrupt software that you thought was using a separate API budget. Source: key types and quota.

Vibe Code Web also has its own lifecycle limits: 100 sessions per day for paid users, a 24-hour maximum session and three-hour inactivity timeout while waiting for a reply. Concurrency depends on the plan. These session counts do not remove token budgets or fair usage. Source: Web session limits.

Which Mistral Purchase Fits Your Buyer?

Choose the API model for the app, the seat for the person, and the deployment boundary for the company. The decision turns on accepted-task cost, the interfaces and organization controls you need, and where eligible data is processed.

Solo builder: start a bounded app on Small 4 and compare its corrections with Medium. The worked month is $13.50 gross on Small. Add Pro when terminal/editor coding or other paid product features justify $14.99 for your own work. The API workload and your coding seat remain separate budget lines, even when eligible bundled credits offset some usage.

Team that codes: choose Team when domain verification, knowledge connectors, per-user storage and data export matter. Start with Pro seats if the organization features do not matter and the account arrangement permits it. Require Enterprise when audit logs or SAML single sign-on are mandatory. Vibe Code Web sessions are personal; collaboration on their output happens through the resulting GitHub branches or pull requests. Sources: plan controls, session sharing.

Company that needs EU hosting: use the regional API when the requirement is eligible inference processing in the chosen geography; request an Enterprise quote when it is private deployment and wider control. On the worked Medium month, the 10% regional uplift makes $150 become $165, if the model is available there. The docs identify api.eu.mistral.ai as the EU endpoint but describe its infrastructure geography as EU and EFTA countries. If your requirement is a particular country or EU-only processing, obtain the exact applicable scope rather than assuming the hostname settles it. Source: regional geography and surcharge.

Regional endpoints currently exclude stateful Agents, Batch and the Files API; only function calling is supported among the documented tools, and model availability varies by region. A document agent dependent on those excluded services cannot be moved by changing a hostname alone. Establish the deployment design before treating the cheaper global API bill as the company's final price. Source: regional restrictions.

Mistral API Docs: Check the Model and Limits

Next week, price a representative accepted task before committing to more seats or a different model. The rate sheet supplies the unit costs; your account and workflow supply the missing volumes and limits.

1. 1

   Record the billing path

   Confirm the application uses a Studio/API key, identify the exact model and service mode, and record fresh input, cached input and output across every agent turn. Keep tool and specialist charges separate.

2. 2

   Check the account and geography

   Read Admin Limits for the chosen model and inspect the available credit balance. For regional traffic, check model availability and feature support, then record the endpoint hostname with each request.

3. 3

   Compare completed work

   Run representative tasks through your own acceptance checks. Count retries and corrections alongside billed usage. Upgrade Small to Medium, Pro to Team, or regional API to private deployment only when the specific result or requirement earns the extra cost.

Frequently Asked Questions

Is Mistral better than GPT?

The prices do not establish that. Medium 3.5's fixed-volume bill is lower than the selected GPT-6.1 Sol short-context bill, but the right choice depends on accepted answers, retries and the workflow's requirements. This article makes no unsupported benchmark ranking.

How much do 1000 tokens cost?

For Small 4, 1,000 input tokens cost $0.00015 and 1,000 output tokens cost $0.0006, calculated by dividing its million-token rates by 1,000. Input and output are billed separately, so a mixed workload needs both counts.

Is Mistral AI free to use?

Yes, the Free product plan includes limited use and advertises $10/month in API credits. The generative API rows are metered; Moderation 2 is specifically marked free. Exact API throughput limits are shown in your account.

Is Mistral AI making money?

Mistral sells subscriptions, metered API usage and custom enterprise deployments. Those published offers describe revenue channels; they do not establish profitability or a current revenue figure.

Who are Mistral AI's main competitors?

For this builder's model shortlist, OpenAI, Anthropic and Google are the compared makers. The table uses GPT-6.1 Sol, Claude Sonnet 5.5 and Gemini 3.8 Flash with prices from each maker's own page.

Is Mistral AI really good?

It belongs on the shortlist when its cost, coding/product features or deployment options fit the job. Evaluate it against your own acceptance checks. A low token price is useful only when the required work succeeds.

What is Mistral AI best used for?

Mistral recommends Medium for most tasks and coding, Small for cost-sensitive projects, OCR for documents and Voxtral for audio. The buyer rule here is to start with the cheaper model that passes the workload's checks and upgrade when corrections justify it. Source: Mistral recommendations.

What happened to Mistral AI?

Le Chat became Vibe, with existing accounts and plans preserved. Current documentation merges chat and work into one experience alongside Code. The current API sheet also marks Large 4 at a sale price, but does not provide a full historical price series or a sale expiry date. Sources: Vibe overview, current prices.

Is Mistral AI a US company?

Mistral identifies itself as headquartered in France with a global presence. That corporate fact is separate from the processing location of a particular API request. Source: Mistral's company description.

Are there discounts for students?

The pricing page lists Pro Education at $5.99/month, with $15/month in API credits, for eligible verified students. Its note limits the offer to twelve months and new Vibe/Le Chat users at accredited higher institutions. Taxes are extra.

Does Mistral Vibe have any usage limits?

Yes. Product usage is subject to plan allowances and fair usage. Code Web currently documents two daily sessions for Free users where enabled and 100 for paid users, with session duration and inactivity limits. These are not promises of unlimited token usage.

Can I get a refund after canceling?

For web purchases, Mistral's policy provides an automatic refund when you cancel within fourteen days of the initial purchase. Later cancellation does not trigger a refund; renewals do not receive that withdrawal window. App-store and carrier purchases follow their channel's process. Source: refund policy.

Can I get a refund for API credits or wallet balance?

Mistral reviews API-credit refunds case by case through support. They are separate from subscription refunds, so do not apply the subscription withdrawal rule to a prepaid balance. Source: credit refunds.

Choose tools around the work they must finish. Get the AI tools map for business owners when you join the newsletter.