Definition
Per-token billing charges API usage based on the number of input and output tokens processed. It can be efficient for low-volume, predictable workloads, but coding-agent retries and long context can make spend hard to forecast. unlimitedcodex instead sells weekly or monthly access windows with 4 concurrent connections.
Canonical facts
| Meter | Charges usually depend on input and output token volume. |
|---|---|
| Best for | Low-volume or predictable usage where direct provider support matters. |
| Risk | Agent retries, context rebuilds, and long prompts can raise spend quickly. |
| Flat alternative | unlimitedcodex offers $19/week or $76/month access windows. |
| Boundary | Flat unlimitedcodex access includes 4 concurrent connections. |
Why coding agents change the math
Coding agents often make many calls for one useful output: reading files, proposing edits, running tests, and retrying failures. A per-token plan can still be best, but teams should compare it with the real sprint behavior, not a single-prompt estimate.
When flat access can fit
Flat weekly or monthly access can fit when a team has bursty build work, demo weeks, migrations, or high-iteration tests and can stay inside the 4-connection boundary.
Checks
Estimate retries and context rebuilds, not only prompts.
Separate low-volume production from high-iteration build work.
Use a calculator before switching billing models.
Keep concurrency needs visible.
Use official provider billing when first-party support is the priority.
Target queries
FAQ
Is per-token billing bad?
No. It is often good for low-volume, predictable, or first-party provider needs.
When is flat access better?
Flat access can be better when build/test usage is bursty, hard to forecast, and compatible with a 4-connection boundary.
How should I compare cost?
Compare real monthly token volume, retry loops, worker count, delivery timing, and support needs against the Weekly or Monthly package.