Work out what it costs, with the assumptions in the open
This is the one page here that shows estimates. It is allowed to, because every number behind them is on screen and you can move it.
Or just compare the subscriptionsWhat are you costing?
Describe it
Categories differ in how many tokens one task takes; Hard debugging is assumed the heaviest. The baselines are published on the methodology page.
The plan
Everything below is an estimate built on the assumptions above, shown as a range rather than a figure.
Your currency is set to USD, so this ranking uses vendors' international prices and leaves out 16 plans priced separately for mainland China.
The link carries every input, including the assumptions. Nothing is stored and no account is needed.
The split of work is inferred automatically from public benchmarks, not endorsed by a person. Where no benchmark covers a category, the choice rests on a general capability tier and says so. Wasted developer time on a failed attempt is not in this model.
Full methodologyWhat the default estimate assumes
One coding turn is taken as 12,000 input tokens, 70% of them served from cache, and 1,500 tokens generated. A developer coding full time runs 20 turns an hour, 6 hours a day, 22 days a month — about 2,640 turns a month. Multiply that by a model's published rate and you have the API route; the subscription route is a plan's monthly price against the allowance it publishes. Every one of these numbers is editable in the panel above, and the result moves with it.
Where these assumptions come fromQuestions about AI coding costs
How much does AI coding cost per developer per month?▾
At the defaults above, one full-time developer moves about 31.7M input tokens and 3.96M output tokens a month. What that costs depends entirely on the model, which is why this page asks rather than answers: the same workload runs to a few dollars on a cheap model and several hundred on a frontier one. The subscription route is flatter — the 55 priced coding plans tracked here run from $3 to $295.44 a month.
What counts as a “turn”?▾
One exchange with a coding agent: your instruction plus everything it reads to answer, then what it writes back. The default assumes 12,000 tokens go in, 70% of them already cached from earlier in the session, and 1,500 tokens come out. It is the unit here because it is what scales with the way you work — hours in the editor, not features shipped.
Does this compare a subscription against API pricing?▾
Yes, and that is the point of it. A subscription is a fixed monthly fee against an allowance the vendor sets; API access is billed per token with no ceiling. The planner prices the same workload both ways and shows where they cross, which is rarely where either vendor's own page suggests.
Where do the prices come from?▾
API rates come from the 362 models tracked here across 14 first-party providers, re-read from each vendor's own pricing page. Plan prices and allowances come from the vendors' pricing and documentation pages, and every one of them carries a link back to the sentence it came from.
How accurate are these figures?▾
They are estimates, and this is the only page here allowed to show them — everywhere else this site refuses to print a number the vendor has not published. The exception is earned by the panel: every assumption behind the result is on screen and adjustable, so a figure you disagree with is one you can correct rather than one you have to take on trust.
Is the calculator free?▾
Yes. No account, no sign-up, and nothing you type is stored — the inputs never leave the page except in a link you choose to share.
Can I share a costing?▾
The inputs live in the address bar, so copying the URL copies the whole scenario — the same team size, models and assumptions, open for whoever you send it to.