Skip to main content

How is a model priced, and why did my task cost more than I expected?

How model prices work, where to check a price before and after a run, why a run can cost more than the listed price, and what isn't charged.

Each model on Wiro has its own price and billing method. You pay after a task finishes, normally only if it succeeded (exceptions below). A listed price is often per unit (per output, per second) or, when it starts with "From", the cheapest option, so your settings can make a run cost more.

How models are priced

  • Per request: a fixed price for each run.

  • Per output: each returned file is charged, or the price once if a successful run returns none.

  • Per second: the model's rate times processing time, from when a server picks up your task until the model finishes.

  • Per token: input plus output tokens, for text models.

  • Per pixel: the price rises with each 1024x1024 (about 1 megapixel) tier of the output image.

  • Per audio second, per character, per video second: by the length of the input audio, the prompt or the output video.

  • Per realtime turn: for live voice sessions.

  • Model-reported: the model reports its own cost when it finishes.

Many models have several price options: in the docs' video pricing example, 720p for 5 seconds costs $0.13 and 1080p for 10 seconds $0.58. The option matching your run's values applies; if none matches, the run is billed per second.

Check the price before you run

  • Price box: under the Run button on a model page, it prices your chosen options, for example Your request will cost ... per output, or says Select the required model options to calculate an estimate. Click View pricing options, if shown, for every option.

  • Per-second models: the box shows the rate per second and sometimes an approximate total.

  • Pricing line: under API quick start, for example Pricing: $0.039/output on google/nano-banana, or a note such as token prices. "From" marks the cheapest option.

  • Pricing page: on the wiro.ai pricing page, enter an amount under YOUR BUDGET to see what it buys per model. Models with several options are counted at the cheapest (marked ≤); click a row for each option. For volume, multiply one run's cost by your number of runs.

  • API: Tool/Detail returns dynamicprice, each option with inputs, price and priceMethod. Without it, a model is billed per second at cps (quegroupcps when cps is 0), and approximatelycost estimates one run (0 with no history yet). See Fallback Pricing in our docs.

See what a run cost

  • In Task History, the Cost column is rounded to three decimals, so a very cheap run can show $0.000. Click a row, then Request parameters to see the options the run used.

  • API: totalcost in Task/Detail. See Billing & Cost in our Tasks docs.

  • Totals and every charge: Project Usage shows your totals, and Transactions lists what each charge took from your balance. See How do I see how much I've spent?

Why a run can cost more than expected

  • A pricier option matched. Higher resolutions, durations or sizes can cost more. Compare Request parameters with View pricing options.

  • A default value applied. Inputs you didn't set use their defaults, and the price follows them. See sending inputs and options.

  • No price option matched, so the run was billed per second. Use values listed under View pricing options.

  • More or bigger outputs. Extra files on per-output models, extra megapixel tiers on per-pixel models and, on some, each input image.

  • Longer inputs or outputs. More audio, prompt characters, video seconds or tokens. Check chat models' price notes: on claude/fable-5, thinking tokens are billed as output and web search separately.

  • Longer processing. Per-second models charge the whole Completed in time in Task History (Generated In on smaller screens).

  • Re-runs. Each is a new task, charged if it succeeds.

What you're not charged for

Wiro charges a task only if the model finished without an error (in the API, pexit is "0"). A failed task's cost is set to $0, and queue time is free. The exceptions are realtime voice sessions, charged per turn, and the Wiro Studio chat, which uses a small amount of credit even when a generation fails. See failed tasks and when you're charged, cancelled and killed tasks, and how to find out why a task failed.

If a charge looks wrong

Send us the task ID (the number after # in Task History) and the price you expected, through the chat on wiro.ai or at [email protected]. API users can send the task token instead.

Related Articles
Did this answer your question?