How to Get a Free Mistral AI API Key (2026)
8 free models available — no credit card required. Get your Mistral AI API key →
Overview
Mistral's Experiment plan — ~1B tokens/month, no card needed.
Mistral AI offers a free "Experiment" plan that provides access to all Mistral open and proprietary models including Mistral Small, Medium, Large, and Codestral. The free tier allows approximately 1 billion tokens/month with a limit of ~1 request/second. Phone number verification is required. Data may be used for model improvement unless opted out.
- Access to Mistral Small, Medium, Large
- ~1B tokens/month free
- Codestral (code model) included
- No credit card required (phone verification)
API Compatibility: OpenAI SDK-compatible (Chat Completions)
Quick Start Guide
- 1 Sign up at console.mistral.ai Phone verification required. No credit card.
- 2 Go to API Keys page
- 3 Create a new API key
- 4 Choose a model Codestral for coding, Mistral Small for general chat.
- 5 Configure OpenAI client Base URL: https://api.mistral.ai/v1
All Free Mistral AI Models — Context Windows & Rate Limits
| Model | Context | Max Output | Modality | Rate Limit | Released | Status | |
|---|---|---|---|---|---|---|---|
| Mistral Small 4 | 256K | 256K | ~1 RPS, 500K TPM | — | Online | Details | |
| Mistral Medium 3 | 128K | 128K | ~1 RPS, 500K TPM | — | Online | Details | |
| Mistral Large 3 | 256K | 256K | ~1 RPS, 500K TPM | — | Online | Details | |
| Mistral Nemo (12B) | 128K | 128K | ~1 RPS, 500K TPM | — | Online | Details | |
| Codestral | 256K | 256K | ~1 RPS, 500K TPM | — | Online | Details | |
| Pixtral Large | 128K | 128K | ~1 RPS, 500K TPM | — | Online | Details | |
| Mistral 7B | 33K | 0 | See provider page | — | Online | Details | |
| Mixtral 8x7B | 33K | 0 | See provider page | — | Online | Details |
Pricing & Limits
Credit Card Not required
Free Tier Permanently free
Context Range 33K – 256K
Total Models 8 free
Rate Limits ~1 RPS, 500K TPM
API Compatibility OpenAI SDK-compatible (Chat Completions)
Use Cases
What Mistral AI's free models are best for, based on aggregated model capabilities:
Limitations & Caveats
- ~1 RPS is slow for concurrent or batch workloads
- Phone verification required for signup
- Free tier data may be used for model improvement unless opted out