← Pricing & setup guides

Monthly API budget or pay per token? Choose the bill you can live with

A subscription price, an included allowance and a usage window describe different things. Compare all three before deciding an AI coding plan is predictable.

The buying decision

Use metered billing for light or irregular work when it is cheaper. Use a finite monthly plan when you value a known subscription charge and its allowance fits your workload.

Four billing models that sound similar

Pay-per-token charges for measured usage. Prepaid credits put money into a balance that requests consume. A monthly API subscription includes an allowance for a recurring fee. A windowed subscription limits usage over periods such as hours or weeks. A product can combine these models, so the presence of a monthly price alone does not establish a hard spending ceiling.

Ask what happens at the limit

Look for the actual exhaustion behavior: requests stop, speed changes, credits are automatically purchased, or overages accrue. Also check whether unused allowance rolls over and whether a team shares a pool. Kilo's pricing, for example, separates platform, inference and hosted compute; OpenCode offers several provider and subscription routes. Add every applicable component before comparing totals.

Compare your quiet and busy months

Use two real months if you have them. Record the settled bill, accepted work and any unused prepaid balance. For a proposed subscription, compare its allowance with representative task consumption rather than assuming that one dollar of allowance equals one dollar at every competing API. Taxes, currency conversion and independent hosted-tool fees can also sit outside the displayed subscription price.

Keep background tasks inside the budget

Scheduled jobs need task limits as well as account limits. Set a maximum run count and a stopping condition, then inspect failures that cause repeated work. For an initial week, record requests from child agents and retries alongside the parent job. A provider's budget stop bounds that provider's included use; it does not stop unrelated cloud-compute fees or requests through a second API account.

Run your own comparison

Use the same task, starting revision and acceptance check. Record every retry and failed attempt so the result reflects the work you actually paid for.

Sources and setup references