Mistral AI’s free tier isn’t a one-time coupon. It’s ongoing, rate-limited access to La Plateforme with roughly 1 request/second, about 500K tokens/minute, and up to about 1B tokens/month per model.
Solo developers shipping a prototype, startup teams pressure-testing model quality, researchers benchmarking prompts and retrieval setups. The Mistral AI free tier is genuinely useful for all of that, as long as you can live with the caps.
This guide covers eligibility, the exact signup flow in the console, what’s included (and what’s gated), and a few practical ways to stretch the limits.
Program at a Glance
| Provider | Mistral AI |
| Credit Amount | Rate-limited free usage (not dollar credits) |
| Duration | Active while on Experiment plan |
| Eligibility | Verified email and phone number; one phone per plan |
| Credit Card Required? | No (Scale upgrade requires one) |
| Difficulty | Moderate. A few console steps plus phone verification. |
| Best For | Prototyping, model evaluation, prompt tests |
| Official Page | Mistral AI Program Page |
What You Actually Get
Mistral’s Experiment plan on La Plateforme is a serverless API plan that provides access to Mistral’s full model lineup for experimentation and prototyping. You can generate chat completions, create embeddings, use function calling, run JSON mode, and call multimodal/vision models where supported. It also includes developer-friendly features like code completion (FIM) with Codestral, document uploads, code interpreter, and “Flash Answers.” The core tradeoff is simple: you pay $0, and you hit rate limits instead of getting billed.
In practical terms, this is enough to build a working demo with real users, validate whether a model fits your task, and do plenty of offline evaluation. If you’re doing heavy concurrency, batch jobs, or high-throughput production traffic, you’ll run into HTTP 429s fast.
Who Qualifies (and Who Doesn’t)
Eligibility is pretty friendly. The Experiment plan is designed for developers who want to try the API without entering payment details, but you do need a verified email and a verified phone number.
- You must sign up on console.mistral.ai and verify your email address.
- A valid phone number is required, because the plan is activated after SMS verification.
- Only one phone number can be used per Experiment plan, so you cannot reuse the same number across multiple accounts.
- No credit card is needed at any point for the free plan (card details come in only if you upgrade to Scale).
If you’re trying to spin up multiple free accounts with the same phone number, you will get blocked. Also, if you can’t accept the Terms of Service and Privacy Policy, you can’t subscribe to the Experiment plan.
How to Sign Up
Plan on about 10 minutes, mostly because of email and SMS verification.
- Go to console.mistral.ai.
- Click “Sign up” and register with email/password, Google, or GitHub.
- Verify your email address using the confirmation link sent to your inbox.
- Once logged in, open the Admin Console from the main menu in Mistral AI Studio.
- Navigate to Subscription and select the Mistral AI Studio tab.
- Click “Compare plans,” then click “Experiment for free.”
- Review the plan details, accept the Terms of Service and Privacy Policy, then click Subscribe.
- Verify your phone number by entering it and confirming the SMS code.
- After activation, go to API Keys at console.mistral.ai/home and click “Create new secret key.”
Once the Experiment plan is active, you can send API requests immediately. If you hit a wall during signup, the most common issue is phone-number reuse across accounts.
What the Credits Cover
Even though people search “how to get Mistral AI credits,” the Experiment plan is closer to “free access with strict caps.” You get the same inference engine as paid tiers, and you can call the full model lineup, but concurrency and usage ceilings are lower.
| Service / Feature | What It Does | Included? |
|---|---|---|
| Chat completions API | Core text generation and chat-style responses. | ✓ |
| Embeddings API | Vector embeddings for search and RAG pipelines. | ✓ |
| Vision (multimodal) | Image + text understanding (supported models only). | ✓ |
| Agents API | Agent workflows and tooling via API. | Partial |
Notable exclusions are the bigger “platform” capabilities: fine-tuning and the Batch API require the Scale plan, and full Agents API access also requires Scale.
Limitations to Know About
Every free program has catches. With Mistral’s Experiment plan, the limits are both technical (rate caps) and policy-related (data usage defaults).
- Rate limits are set per model at the workspace level, and they’re meant for testing rather than production.
- You should expect roughly 1 request per second per model, so parallel workloads will throttle quickly.
- Token limits include both input and output tokens combined, which matters for long-context prompts.
- If you exceed the caps, you receive HTTP 429 errors and requests are not queued.
When you hit the limit, you don’t get billed, and there’s no surprise overage charge. Your calls just fail until you slow down, switch models, or upgrade to the Scale plan (which requires adding a credit card and moves you to token-based billing).
Have Unused Mistral AI Credits?
Not everyone is on the free plan forever. Teams that upgrade to paid tiers or sign larger agreements sometimes end up with usage they can’t burn down before internal timelines change. Honestly, unused credits are common. If you have surplus Mistral AI credits you won’t use, AI Credit Mart lets you sell them instead of watching them expire with zero value.
Need More Mistral AI Credits?
If the Experiment plan feels tight (it will), the next step is usually Scale with pay-as-you-go billing. You don’t always have to pay list price, though. AI Credit Mart has discounted Mistral AI credits from orgs with surplus allocations, typically 30–70% below retail, which can stretch a startup budget a lot further.
Tips for Getting the Most Out of Your Credits
- Opt out of training data collection if your prompts contain anything sensitive, because Experiment uses inputs/outputs for training by default.
- Check your live limits in the Admin Console at admin.mistral.ai/plateforme/limits since exact caps vary by model and can change.
- Plan for HTTP 429s by adding backoff and retry logic, because requests are not queued when you hit rate limits.
- Use OpenAI-compatible client libraries and just swap the base URL and API key, which saves you time when prototyping.
- Remember Le Chat and La Plateforme are separate products with separate free tiers, so don’t troubleshoot the wrong account.
Frequently Asked Questions
They’re not dollar credits. You get free usage at standard per-token pricing (so $0 billed), but you’re constrained by per-model rate limits like ~1 RPS, ~500K tokens/minute, and up to ~1B tokens/month.
No. The Experiment plan requires a verified phone number, not a credit card.
There’s no stated expiration window for Experiment because it’s a free plan with rate limits, not a time-boxed credit grant. As long as you stay on the Experiment plan, you keep the same general access. If Mistral changes limits or plan terms, your current caps may change (they recommend checking your live limits in the admin dashboard).
Yes. If you have Mistral AI credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.
AI Credit Mart has discounted Mistral AI credits available from companies with surplus allocations. Prices are typically 30-70% below retail.
The Experiment plan doesn’t “expire credits,” because it isn’t a credit balance. If you’re on paid credits from another source, expiration depends on that specific agreement, and you should verify the terms in your contract or dashboard.
Yes, by default. You can opt out in the Admin Console by going to Privacy and disabling “Anonymous improvement data.” On paid plans like Scale, data is opted out by default. If you’re testing with real customer text, don’t skip this step.
Slow down and retry with backoff, because HTTP 429 requests are not queued. You can also distribute traffic across different models since limits are per model. If you’ve outgrown the plan, upgrade to Scale for higher rate limits (credit card required), and your API keys and endpoints stay the same so the switch is usually painless.
Mistral’s Experiment plan is one of the better free API tiers right now: full model access, no card, real prototyping headroom. Use it to validate fast, and if you ever end up with surplus paid credits, you can sell them instead of wasting them.
Your AI credits are losing value every day
Join the marketplace and start trading unused credits today.