A prepaid GPT API for testing and production workloads.

Current sale rates

Prices stay listed while a model is temporarily unavailable. Only models marked LIVE can be called.

How to compare GPT API prices honestly

A per-token price is only half of the bill. Count three things for your own workload: fresh input, cached input and output tokens. Coding agents re-send the same long context on every step, so most of their input is cached and the cached rate decides the total.

Model.sale charges one flat rate per million charged tokens and counts cached input at 40%. For a session: charged tokens = fresh input + output + 0.4 x cached input. Multiply by the model rate from the table above. The pricing methodology shows the same calculation against the vendors' list prices.

When this is cheaper, and when it is not

• Cheaper for cache-heavy agent work such as Codex CLI and Cursor sessions, where most input is repeated context.
• Cheaper for plain chat with long answers, because the flat rate is far below the list price of the larger models.
• Close to list price for the smallest models. Those are marked in the catalog and are not sold as discounts.

Prepaid also means a hard ceiling: the wallet cannot go below zero, and each key can carry a daily spend limit.

Start in three steps

1. Try the free demo on the home page (three questions a day, no signup).
2. Create an API key and set the base URL to https://api.model.sale/v1 in your OpenAI client.
3. Add balance from $5 only when you need it and watch the exact charge of every request in the dashboard.