Mistral Experiment Plan Data Privacy: What Free Tier Opt-In Means for Your Prompts

Opting into Mistral's Experiment plan means you explicitly consent to having your API request and response payloads collected, stored, and used to train future models, effectively waiving privacy guarantees for that data.

The cheahjs/free-llm-api-resources repository tracks free LLM API offerings and documents that Mistral's La Plateforme Experiment plan requires users to accept data training terms. Understanding Mistral Experiment plan data privacy implications is essential before transmitting sensitive information through the free tier.

What Is the Mistral Experiment Plan?

The Experiment plan is Mistral's free-tier offering on La Plateforme. According to the repository's generated documentation at README.md lines 111-116, this tier "requires opting into data training".

This opt-in represents a contractual exchange: you receive free API access in return for granting Mistral rights to use your traffic for model improvement.

Data Collection and Usage Terms

When you enable the Experiment plan, you authorize Mistral to treat your prompts as training material. The free-llm-api-resources documentation outlines specific data handling practices that apply to this tier.

Request and Response Logging

Under the Experiment plan, every input prompt and generated output is logged on Mistral's servers. This includes full request/response payloads transmitted to model endpoints such as https://api.mistral.ai/v1/chat/completions.

Training Use and Retention

Logged data may be incorporated into future model training runs, fine-tuning processes, or evaluation datasets. The provider does not specify a fixed retention period; data is retained at least as long as necessary for training purposes.

Privacy and Compliance Implications

The Experiment plan offers no explicit confidentiality promise. Your data is treated as non-private and may be examined by Mistral staff. For organizations requiring GDPR compliance or strict privacy controls, this plan is technically unsuitable because the opt-in constitutes explicit consent to broad data reuse.

Source Code Context

The privacy requirements are hardcoded in the repository's data generation logic. In src/pull_available_models.py, the script generates the README content including the specific line stating that the "Free tier (Experiment plan) requires opting into data training".

The file src/data.py enumerates the specific Mistral model identifiers—such as mistral-7b-instruct—that can be queried under these terms. These implementation files confirm that the training requirement is an inherent property of the free tier rather than an optional configuration.

Using the Mistral API: Privacy Considerations

The API interaction remains technically identical between plans; the privacy implication changes based on your account status.

Basic API Call (Experiment Plan)

This Python example demonstrates calling a Mistral model. If your account uses the Experiment plan, this request will be logged for training:

import requests
import os

MISTRAL_ENDPOINT = "https://api.mistral.ai/v1/chat/completions"
API_KEY = os.getenv("MISTRAL_API_KEY")

payload = {
    "model": "mistral-7b-instruct",
    "messages": [{"role": "user", "content": "Explain quantum computing"}]
}

headers = {
    "Authorization": f"Bearer {API_KEY}",
    "Content-Type": "application/json"
}

response = requests.post(MISTRAL_ENDPOINT, json=payload, headers=headers)
print(response.json())

The repository does not store API keys; you must configure MISTRAL_API_KEY in your environment variables.

Disabling Training (Paid Plans Only)

Some paid plans offer metadata flags to disable training. While the repository does not provide a wrapper for this functionality, the standard implementation would be:

payload["metadata"] = {"disable_training": True}

Note: This flag only functions on paid tiers that explicitly offer non-training options. The Experiment plan does not respect this flag.

Alternatives for Privacy-Conscious Users

If you require strict data privacy, the repository suggests evaluating other providers. Options like OpenRouter's free tier or OpenCode Zen's free models may offer different privacy terms that do not require training opt-in, as noted in README.md lines 14-22.

Summary

• Mistral Experiment plan data privacy requires accepting that all prompts and responses become training data.
• The opt-in requirement is documented in the free-llm-api-resources repository's README and generation scripts at src/pull_available_models.py.
• Data is logged persistently with no guaranteed retention limit and no confidentiality promises.
• The Experiment plan is incompatible with GDPR-strict use cases due to the explicit consent to data reuse.
• Paid plans may offer disable_training flags, but the free tier does not support privacy-preserving API calls.

Frequently Asked Questions

Does Mistral store my prompts forever under the Experiment plan?

The provider does not publicly disclose a fixed retention period for Experiment plan data. According to the repository documentation, data is retained at least as long as needed for training purposes, which implies indefinite storage until it is either incorporated into models or purged according to internal schedules.

Can I use the Mistral API without opting into data training?

No. The free Experiment plan specifically requires opting into data training as a condition of access. To avoid training use, you must either upgrade to a paid plan that offers non-training options or select a different provider from the free-llm-api-resources list that does not require such opt-in.

Is the Experiment plan GDPR compliant?

The Experiment plan is generally unsuitable for GDPR-strict requirements because you must explicitly consent to data reuse and training. While GDPR allows processing with consent, the lack of specific retention limits and the broad use rights granted to Mistral make it difficult to guarantee data subject rights like the right to erasure or processing limitation.

How do I verify if my account is on the Experiment plan?

Check your Mistral La Plateforme dashboard account settings. The repository's documentation indicates that the Experiment plan is the default free tier, so if you have not entered payment information or explicitly selected a paid tier, you are likely subject to the training opt-in terms documented in the README.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}