Loading the calculator.
Loading the calculator.
Lab
What a month of coding-agent use costs if you pay per token through the API, next to the flat monthly plans.
API bill $48.80 for 80 sessions on Claude Opus 5.5. The cheapest plan is more than 10% cheaper than paying per token.
Plans have usage limits this calculator can’t see. If you hit the limit and keep going on extra credits or the API, the plan stops being a flat price.
Still Building
A token is a small chunk of text. The API charges per million tokens, with one rate for what the model reads (input), a lower rate for input it has already seen and kept in a cache, and a higher rate for what it writes (output).
Sessions are working days times sessions per day. The cached share of input is billed at the cache rate. The rest of the input is billed at the full input rate. Output is billed on its own. The API bill is those three added up for the month.
The comparison uses the cheapest plan from the same vendor. The API wins when its bill is under 90% of that plan’s price. The plan wins when the API bill is over 110%. In between, they are about equal.
Every rate below was read on the vendor's own pricing page on 2026-09-26. Claude rates use the standard tier, not batch or fast mode. OpenAI's faster gpt-5.3-codex tier costs twice the standard row and is left out. Saving text to the cache isn’t counted. That costs more than a plain input token, so a session that starts with a lot of text comes out slightly cheaper here than it really is.