How to get a Mistral API key, free plan included

Getting a Mistral API key takes five minutes on La Plateforme, the developer console of the French provider. You create an account, pick a plan and click "Create new key". Mistral also offers a free plan called Experiment, available after a simple SMS check, so you can try the models with no credit card before moving to pay-as-you-go billing if you need it.

Two things set Mistral apart from the other providers. Its prices sit among the lowest on the market and it's a European company subject to EU law, which matters as soon as you handle client data. The prices quoted were recorded on October 16, 2026 from the official price list. Every figure can be checked in two clicks from the sources listed at the bottom.

01Key takeaways

• You create the key on console.mistral.ai ("La Plateforme") with an email address or a Google or GitHub account.
• The free Experiment plan opens the models after an SMS check, with no credit card. Its throughput is deliberately capped, which is fine for testing and too tight for production.
• On pay-as-you-go, the October 16, 2026 price list runs from $0.15 per million input tokens (Mistral Small 4) to $1.50 (Mistral Medium 3.5). That puts Mistral among the lowest prices on the market.
• The key is shown only once and can get an expiration date at creation. Copy it straight to the right place.
• Mistral is a French company based in the European Union, a strong argument for professional uses where GDPR matters.

02Why Mistral deserves a spot on your key ring

In the logic of AI without a subscription, each key is an engine you plug in for a given task. Mistral holds two positions that no other provider covers as well. The first is price. Its recent models are listed at rates that make it one of the cheapest providers on the market for comparable quality, which adds up fast on large volumes.

The second is location. Mistral is a French company, its models are served from a European platform and EU law applies. For an agency handling client content, for a regulated sector or simply out of a preference for sovereignty, that's a criterion you can defend in a meeting without a single line of technical explanation.

One last practical advantage is that the Mistral API is compatible with the OpenAI format. Most BYOK tools therefore support it without any adaptation. You just paste the key in the same place as the others and pick Mistral from the provider list.

03Step 1. Create your account on La Plateforme

Go to console.mistral.ai, the Mistral developer space, which is separate from the consumer chat interface (le Chat). Sign-up accepts a regular email address or a Google or GitHub account. As with the other providers, the console account has its own billing, independent of any chat subscription.

When you create the account, the console asks you to pick a plan. That's where you choose between the free trial and billed usage, as explained in the next step. Nothing is irreversible, since you can switch from one to the other in the settings at any time.

La Plateforme then organizes work into workspaces, each with its own keys and billing. If you work alone, you won't touch them. In an agency, they're what lets you separate clients cleanly. That habit makes billing at actual cost easier and keeps revocation simple the day a contract ends.

04Step 2. Free Experiment plan or pay-as-you-go

The Experiment plan is a welcome quirk of Mistral. It gives free access to the models, validated by an SMS check, with no payment method. In return, throughput is heavily limited because the plan is built for experimenting. The exact conditions appear in the console when you choose it. It's ideal for testing model quality on your real tasks, but it will quickly feel tight once you connect a tool in production.

The pay-as-you-go plan works like everywhere else, with a payment method, billing per million tokens and adjustable caps. If your priority right now is to pay nothing, the comparison of free API key offers sets Experiment against the Google free tier. If you only want to try AI without a card, both will do the job.

You switch between the two in the billing settings, without recreating an account or a key. You start on Experiment to check quality on your own tasks, then move to the paid plan the day the capped throughput starts to bite. That usually happens when you connect a tool that sends requests one after another. This smooth progression is the spirit of paying per use, because commitment follows need and never the other way around.

05Step 3. Generate the key

1. In the console, open the API Keys section.
2. Click "Create new key".
3. Name the key after its use. A good idea specific to Mistral is to set an expiration date if the key is for a trial, so it deactivates on its own.
4. Copy the key right away, because it won't be shown in full again.

The optional expiration is worth adopting. A trial key that dies on its own after a month is a forgotten key that will never linger somewhere. Few consoles offer this setting at creation. It's a small ergonomic touch typical of La Plateforme. For production keys, deliberate rotation works better, as explained in the habits that protect your keys.

06The Mistral catalog in two minutes

Before the price list, it helps to know what the catalog covers, because Mistral doesn't only do text. First come the three general-purpose models. Small handles everyday work quickly and cheaply, Medium is for polished writing and Large sits at a surprising middle price point that makes it excellent value for volume.

Around them are the specialists. Codestral handles code, a vision model analyzes images, embedding models power semantic search and there's even an OCR component billed per page to extract text from documents. For a site publisher, the general-purpose models cover almost every need. The specialists wait for the day a specific case comes up. The same key gives access to them with no extra step.

All of them follow clean versioning. A "latest" alias points to the current version while dated versions let you freeze a behavior. In a BYOK tool, the alias is enough in nearly every case. Dated versions remain a matter for custom integrations.

07Mistral prices from the official price list

Prices are listed per million tokens, input first and output second. They were recorded on October 16, 2026 on Mistral's official price list. Batch processing cuts every amount in the table in half.

Mistral API prices per million tokens, official price list of October 16, 2026

Model	Input	Output
Mistral Medium 3.5	$1.50	$7.50
Mistral Large 3	$0.50	$1.50
Mistral Small 4	$0.15	$0.60

One detail jumps out right away. Large 3 is listed below Medium 3.5. Mistral positions its range by use more than by size, so the name doesn't tell you the price. Always read the list rather than the model name. Per unit, a 1,500-word article generated on Mistral Small costs less than one cent, output included, while Medium comes to about two cents. Against the market, these levels put Mistral near the bottom of the big providers' price lists, as the breakdown of AI API prices compared shows. For tight budgets, DeepSeek and the newer budget models of Anthropic and OpenAI compete in the same league. The low-cost DeepSeek key has its own guide, with its own trade-offs on where data is stored.

Model choice follows the usual BYOK rule. Small handles everyday tasks and large volumes, while Medium is for when writing quality comes first. The method to find the cheapest model that meets your standard applies here word for word.

08Three uses where Mistral prices change the math

Large content volume comes first. Take the standard calculation, where 300 articles of 1,500 words represent about 750,000 output tokens. On Mistral Small 4, that comes to 45 cents of output and barely more all in. The same volume billed in credits on a content platform would run into hundreds of euros. It's the widest gap we've found across all the provider guides, worked out from the official price lists.

Client data is the second case. When a contract or a data protection officer (DPO) requires you to justify how data flows, "French provider, EU law" shortens the discussion. It's a meeting argument more than a technical one. In that setting it carries its full weight.

Background automation is the third. Think of summarized news monitoring, rewording in bulk and repetitive tasks. At these price levels, automation stops being a budget line and becomes background noise, especially once you add batch processing, which halves the amounts again.

09Secure and organize your keys

An API key is handled like a password, free or paid. Never publish it or paste it into a code repository or a screenshot. Only give it to tools that store it in your browser or in encrypted form. Revoke it at the slightest doubt. Revocation takes two clicks from the API Keys page and a replacement key can be generated right after. Mistral adds two conveniences. One is the expiration date at creation, mentioned earlier. The other is workspace management, which keeps uses or clients cleanly apart.

For budget, the reflex is the same as everywhere. Set a cap from the start and watch consumption during the first weeks. The orders of magnitude to expect are in the breakdown of what usage really costs. For writing, even heavy writing, Mistral comes to tens of cents a month.

10Connect the key and start producing

Thanks to format compatibility, the Mistral key works in most BYOK tools. At OpenTahr, you paste your Mistral API key into the settings of a free account. It stays in your browser and never reaches our servers. Every tool then works with it. The classic pairing is Mistral Small for volume and a high-end model from another provider for showcase texts. Switching engines takes two clicks per task, which is the whole point of having several keys in the same workshop.

To equip the rest of your key ring, the provider guide hub gives the steps console by console. If you're coming from OpenAI, you'll be on familiar ground. The OpenAI key setup follows the same logic, apart from the menu names.

11Frequently asked questions

Is the Mistral API free?

The Experiment plan allows real free usage, validated by SMS, with throughput capped for testing. Serious usage goes through the paid plan, billed per million tokens with no subscription or monthly minimum, so a month without usage costs zero. At these prices, the line between the two is more psychological than financial. A few cents cover a month of individual production.

What is the difference between le Chat and the Mistral API key?

Le Chat is the consumer chat interface. The API key gives access to the models from your own tools, billed by consumption. The two are managed separately and using one gives you no rights on the other. To produce content with tools, the key is what you need, while le Chat remains a good place to test conversations.

Is Mistral as good as OpenAI or Anthropic for writing?

For everyday writing, the recent Mistral models hold their own very well, at a fraction of the price. On the most demanding texts, opinions differ depending on use. Run the test that settles it. Send the same prompt to both models, then judge the results on your own content.

Does the key work in tools built for OpenAI?

Very often it does, because the Mistral API uses the same format as the OpenAI one, which most tools can speak. You still need to check that the tool lists Mistral among its providers or accepts a custom endpoint. Serious BYOK tools do.

Why choose a European provider?

Compliance comes first. Mistral is an EU company under European law, which simplifies GDPR discussions with clients or a DPO. Resilience comes next, because a European engine next to the American ones in a BYOK key ring means one less dependency. It comes at no extra cost, since the price list is one of the lowest on the market. The sovereign argument costs nothing, which is rare enough to point out.

Transparency note. Prices were recorded again on October 16, 2026 from the official Mistral pages linked below. The Experiment plan conditions date from the June 13, 2026 check and appear in the console, which is the reference. Mistral does not pay us anything.

12Sources

1. Mistral, official API price list
2. Mistral, documentation on tiers and rate limits
3. Mistral, API price list per model (Small 4 at $0.15 and $0.60)