One low-cost AI API key, without a subscription.

Short answer: when is prepaid cheaper?

Prepaid is often a good fit for prototypes, coding agents and irregular workloads where a recurring seat would sit unused. The fair comparison is your own input/output mix, not a headline “discount”: Model.sale publishes a blended charge per 1M charged tokens, while many official APIs split input, cached input and output prices.

A temporary reservation protects your balance before dispatch. Once terminal usage arrives, the exact charge is settled and the unused reservation is released. The selected model is never silently replaced with another one.

Compare live model prices before you integrate.

Rates below are USD per one million charged tokens. Availability is refreshed automatically; a model can keep its price visible while temporarily unavailable, but only a LIVE model accepts a request.

The full catalog, minimum request charge and health status are on Models. There are currently 19 live priced models.

Good fits for a low-cost prepaid API

Three steps to your first request

1. Create an account and verify your email or connected identity.
2. Create a key, set a daily or monthly limit, then add at least $5 on the billing page after sign-in.
3. Copy the base URL and run the minimal example in the API docs. Check Usage for terminal tokens and the exact charge.

This example uses currently published model gpt-5.5 with /v1/chat/completions. Availability can change; confirm the ID in GET /v1/models.

curl https://api.model.sale/v1/chat/completions \
  -H "Authorization: Bearer $MODEL_SALE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.5","messages":[{"role":"user","content":"Say hello in one sentence."}]}'

How to compare with official APIs and aggregators

Official APIs may offer native features, enterprise contracts or separate input/output rates. Aggregators may offer a wider catalog and routing. Model.sale focuses on a simple prepaid wallet, a curated public catalog and request-level usage. Use the comparison page and workload calculator with the same token mix before deciding.

Do not treat the lowest per-million number as a universal saving: cache rules, output share, retries, minimum charges and model behavior change the effective cost.

Questions developers ask