ModelHubby

← All guides · provider record →

Mistral La Plateforme — free tier setup

What you get free

The free "Experiment" plan on La Plateforme: generous per-model throughput across Mistral's lineup (large, medium, small, nemo), no credit card. Catalog quota: "Limits (per-model): 1 request/second, 500,000 tokens/minute, 1,000,000,000 tokens/month (cheahjs README). mnfst data.json: Free 'Experiment' plan, no credit card. ~1B tokens/month." Category: conditional_free — and here is the condition, read it before you sign up:

The free Experiment plan requires opting into data training. Per the catalog: "Free tier (Experiment plan) requires opting into data training"; "Prompts may be used to improve models." Do NOT send customer data, proprietary code, or anything confidential through the free tier. (Sourced from seed repos; not confirmed against an official Mistral page — the help-center and docs tier URLs returned 404 at check time.)

Friction check
Credit card no
Phone verification yes (per cheahjs README)
Org/membership no
Est. time to first key 10m
Trains on your prompts yes — free plan is conditioned on the data-training opt-in

Get a key

Official step-by-step docs could not be fetched (404s); steps below are the known flow — treat button names as (unconfirmed — verify on first signup).

  1. Go to console.mistral.ai (the docs' "Try Studio" link) and create an account.
  2. Verify your phone number when prompted (required per catalog).
  3. In workspace/billing settings, select the free Experiment plan. This is where the data-training opt-in happens — you are trading prompts-as-training-data for free usage (unconfirmed exact dialog wording).
  4. Open the API Keys section of the console and create a new key (unconfirmed exact menu name).
  5. Copy the key immediately.

First request

Base URL from the catalog; model from the catalog. Success shape: JSON with choices[0].message.content.

export MISTRAL_API_KEY="YOUR_KEY"
curl https://api.mistral.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -d '{"model":"mistral-small-2603","messages":[{"role":"user","content":"Say OK."}],"max_tokens":16}'

Limits that will bite you

  • 1 request/second (expressed under rpm in the catalog) — the hard ceiling for any parallel workload. Queue and serialize.
  • 500,000 tokens/minute and ~1,000,000,000 tokens/month, per model (catalog raw_text). The monthly cap is huge but real; it is per-model, not pooled.
  • These figures are unverified (catalog verification status: unverified, confidence 0.5, imported from seed repos) — probe before depending on them.
  • The data-training condition is the operational limit that matters most: free tier is unsuitable for confidential workloads regardless of throughput.
  • Separate free Codestral endpoint exists (codestral.mistral.ai, 30 RPM / 2,000 RPD, phone verification) but is not modeled in this catalog record.

Key rotation runbook

  • Second key: create another key from the console's API Keys section (multiple keys per workspace expected but unconfirmed — verify on first signup).
  • Revoke: delete the key from the same API Keys section (exact button label unconfirmed).
  • ModelHubby integration: export MISTRAL_API_KEY_2="..." — the key pool picks it up automatically; state machine cools down rate-limited keys per-key.

ModelHubby usage

pnpm mh show mistral
pnpm mh probe --live --provider mistral --apply