Principles

Pay per token. Every model lists its public per-1M price on the Models page. What you see is what we charge — no hidden markup baked into the rates.

No surprise bills. The wallet is prepaid only. Usage is deducted from your balance as it happens, and requests stop when it runs out — never a bill after the fact.

OpenAI-compatible everywhere. One base_url swap moves any existing OpenAI SDK integration to CheapestToken. No new SDKs, no rewrites.

Budgets enforced at the proxy. Keys and spend limits are checked at the gateway on every request, before a single token is billed.

Transparent usage. Per-request logs and daily usage breakdowns let you verify every charge against actual traffic.

Fair cache pricing. Cached prompt reads are billed at a discounted rate and shown as a separate column on the pricing table, not quietly folded into full price.

Every key draws from a wallet you top up in advance. Usage is deducted as it happens and requests stop when the balance runs out — there is never a surprise invoice after the fact.

Keys are revocable at any time from the console, so a retired or leaked key stops billing the moment you revoke it.

We buy committed-use and bulk pricing upstream and pass the discount through. The catalog shows the upstream list price struck through next to the price you actually pay, model by model.

One OpenAI-compatible endpoint serves every model we route — the same key and the same base URL work across the whole catalog.