Cost per task

Not dollars per million tokens — dollars per job.

Sticker price is marketing. Your bill is decided by output tokens, and reasoning models spend several times more of them on the same job. We run the jobs and report medians.

Korean text gets it wrong twice. Token counts differ per model for the same sentence, so identical sticker prices can mean double the real cost. KRW, VAT, tax invoices, tokenizer efficiency →

New to these terms? 30-second primer

Token: The smallest unit an AI reads and writes. Roughly 1-2 tokens per English word; Korean is about 0.3-0.7 tokens per character. Billing is based on this count.
Input / output: Input is what you send; output is what the model writes back. Output is usually several times more expensive.
Reasoning tokens: What the model thinks through before answering. You never see it, but it is billed as output. This is why bills come in far above expectations.
Context: The maximum length of text you can send in one call.
Cache hit: The discount you get when the same leading portion is sent repeatedly.

What are you building?

Use case and scale are enough for a monthly figure

1 · Use case

2 · Scale

3 · Assumptions behind this estimate click to edit

—

incl. 10% Korean VAT—

Margin % → Suggested sell price —

Pick a job

Workloads 0 real-world jobs

No measured data yet Price collection is running, but the benchmark has not executed yet. We do not fill rankings with estimates.