Why look at these three instead of another OpenAI walkthrough

Most "free token" guides orbit OpenAI, Gemini, and Claude. But the genuinely generous free tiers often sit with second-tier vendors: they are competing for developers, so trial quotas are looser, approvals are lighter, and non-US accounts are treated more kindly.

This article covers exactly three: Mistral (France, La Plateforme), Cohere (Canada, Trial Keys), and AI21 (Israel, AI21 Studio). What they share: sign up and go, no corporate paperwork, and free tiers that persist rather than running as limited-time promos.

> Quota numbers shift constantly. This article avoids hard figures and uses "about / typically" throughout. Trust whatever your console shows on signup day.

Steps

Step 1: Mistral La Plateforme — phone verification unlocks the free experiment quota

  1. Open the La Plateforme console on Mistral's site and register with email or Google.
  2. Go to Billing / Plan and confirm the account is in the free experiment state. Mistral typically grants phone-verified new accounts a usable free quota for models like mistral-small and mistral-large.
  3. The critical action is phone verification. Unverified accounts can see the UI but often cannot actually call the API. The quota lands only after verification.
  4. Create a Key under API Keys. The full key is shown only once — store it in a password manager immediately.
  5. Verify with a minimal request:

```bash

curl https://api.mistral.ai/v1/chat/completions \

-H "Authorization: Bearer $MISTRAL_API_KEY" \

-H "Content-Type: application/json" \

-d '{"model":"mistral-small-latest","messages":[{"role":"user","content":"ping"}]}'

```

Requirement: a phone number that can receive SMS; one number usually verifies only one account.

Step 2: Cohere Trial Key — no credit card, but roughly a 90-day clock

  1. Register a Dashboard account on Cohere's site (a personal email works; no corporate domain required).
  2. Open API Keys in the left menu. You will see a key labeled Trial. That is the free-tier entry point: no credit card needed, but it typically expires after about three months, after which you need a Production Key (which does require a card).
  3. Trial Key rate limits are far lower than paid tiers. It is fine for development and small evaluation batches, not for load testing.
  4. Supported models include the command family and the embed family. Free vector quota for RAG or semantic search basically comes through embed.

```bash

curl https://api.cohere.com/v1/chat \

-H "Authorization: Bearer $COHERE_TRIAL_KEY" \

-H "Content-Type: application/json" \

-d '{"model":"command-r","message":"ping"}'

```

Requirement: the Trial Key has a hard countdown from creation day. Add a calendar reminder the moment you create it.

Step 3: AI21 Studio — quota granted at signup, Jamba family

  1. Register at AI21 Studio and generate a Key under API Keys.
  2. AI21 usually grants new accounts a free token quota for its Jamba family. Once it is consumed, you must attach a payment method to continue.
  3. That quota generally has no hard expiry date, but it tracks account activity and policy changes. Treat it as "use it until it is gone."
  4. The API follows an OpenAI-compatible style, so migration cost is low:

```bash

curl https://api.ai21.com/studio/v1/chat/completions \

-H "Authorization: Bearer $AI21_API_KEY" \

-H "Content-Type: application/json" \

-d '{"model":"jamba-1.5-mini","messages":[{"role":"user","content":"ping"}]}'

```

Requirement: the quota is finite and does not refresh. Use it as a backup upstream, not a primary one.

Step 4: Chain the three into one local supply line

Individually none of these is large, but the interfaces are similar enough to sit behind a single local gateway (LiteLLM, or a small custom router), with fallback from small models to larger ones:

```yaml

# litellm config excerpt

model_list:

  • model_name: cheap

litellm_params:

model: mistral/mistral-small-latest

api_key: os.environ/MISTRAL_API_KEY

  • model_name: cheap

litellm_params:

model: cohere/command-r

api_key: os.environ/COHERE_TRIAL_KEY

  • model_name: cheap

litellm_params:

model: ai21/jamba-1.5-mini

api_key: os.environ/AI21_API_KEY

```

With three free upstreams behind one model_name, a request automatically falls through to the next vendor when one runs out of quota or hits a rate limit.

Caveats

  • Numbers move: free quotas, rate limits, and model availability at all three vendors can change at any time. Trust the console on signup day.
  • The Cohere Trial Key expires: this is the easiest trap of the three. Set a reminder.
  • Phone number is Mistral's gate: without a working number you get no real quota, and virtual number ranges may be rejected.
  • Do not run production traffic on free tiers: rate limits, concurrency, and SLA are unsuitable for live services. Development, evaluation, and prototypes only.
  • A leaked key means a drained quota: free keys have no spend cap, so a leak can be burned through by someone else.

When this fits

  • Individual developers who want to compare European, North American, and Middle Eastern models without attaching a card.
  • Small teams doing model selection (A/B evaluation) who want to prove the workflow before paying anyone.
  • RAG projects needing a backup upstream: Cohere's free embed quota plus Mistral's generation is a common pairing.
  • Teaching and demos: no billing involved, and the experiment environment can be reset freely.