We use cookies to keep you signed in. Cookie policy
We use cookies
CalculatorAI uses cookies to keep you signed in and remember your preferences. Optional cookies help us understand how the site is used. Cookie policy
About $0.5574 per task on Claude Sonnet 5: 15 model calls, each resending the conversation so far. At 400 tasks a month that is $222.96.
A coding agent plans, reads files, edits, runs the tests and fixes what broke — around fifteen calls, with file contents and test output flowing back into the conversation on every step. On Claude Sonnet 5, one task costs about $0.5574, and 400 tasks a month come to $222.96. The number that decides it is not the price per million tokens but how many calls one task takes — here 15 — because every call resends the whole conversation so far.
So the calls are not the same size. The first sends about 3,400 input tokens; the last, carrying everything before it, sends about 32,800 — 9.6× more. Long loops are where the square bites: halve the steps and the input bill falls by roughly three quarters, which is why tight tasks and a cached system prompt matter more here than anywhere else. The calculator below opens with this shape loaded: change the steps, the tool output and the volume to your own agent, and treat every figure here as an estimate from those assumptions, not a quote.
What One Agent Task Really Costs
Cost per task
One task · 5 calls to GPT-5.6 Terra
Cost per model call
$0.008640
averaged across 5 calls
Context growth
3.3×
1,800 tokens on the first call, 6,000 on the last
What drives one task
$0.0432At this volume
100 tasks · DayIf you ran this many tasks
100
$4.32
1,000
$43.20
10,000
$432.00
100,000
$4,320.00
What this figure assumes
Drop in a paystub, bank statement or screenshot — the AI reads it and fills these fields for you.
Income & Expense Tracker is where this stops being a one-off number: keep your real balances there and the projection updates itself.
Describe an invoice and the AI writes it. PDF without watermarks, free.
Same task shape on every row (3,000 system tokens cached, 400-token task, 600 tokens written and 1,500 returned by a tool per step). The last column doubles the steps: the input bill roughly quadruples because each call carries all the earlier ones.
| Model | Per task | Per month (400 tasks) | Per task at 30 steps |
|---|---|---|---|
| Claude Sonnet 5 | $0.5574 | $222.96 | $2.05 |
| Claude Opus 5 | $1.39 | $557.40 | $5.14 |
| GPT-5.6 Terra | $0.5754 | $230.16 | $2.09 |
| GPT-5.6 Luna | $0.0575 | $23.02 | $0.2090 |
Use cases