Kingy AI Workload Cost Calculator

Estimate the cost of running an AI workload, including caching, retries and human review. Compare the total with the number of results you can actually use. This calculator works with unexpired starter rates or your custom rates while KAPI-backed current pricing is paused.

Checking available provider prices. Custom rates are always available.

1. Choose your rates Rate source

Model and version Input · $ / 1M tokens Output · $ / 1M tokens Cache read · $ / 1M tokens Cache write · $ / 1M tokens

Custom rates are supplied by you and are not independently verified. Enter the rates for your selected service tier, region and context length. Use 0 only when the price is actually zero. Provider rates are shown below only while their source records pass the calculator’s freshness checks.

2. Describe the workload

Attempted tasks / month Base API calls / task Input tokens / call Billed output tokens / call Extra retry calls / 100 tasks Accepted results · % Cacheable input tokens / call Cache hit rate · %

Retries add calls to attempted tasks; they do not count as new tasks. Output should include any billed reasoning tokens. All calls use the same token estimate.

3. Add the cost of getting work done

Tools and other costs · $ / task Human review · minutes / task Reviewer cost · $ / hour Fixed monthly cost · $

Enable JavaScript to use the calculator. You can also estimate token cost manually: calls × (input tokens × input rate + output tokens × output rate) ÷ 1,000,000, then add review and fixed costs.

How the estimate works

Calls = attempted tasks × (base calls per task + extra retry calls per 100 tasks ÷ 100). The calculator prices ordinary input, cache reads, cache writes and output separately. A cache miss uses the cache-write rate. Human review is charged once per attempted task.

Use measured cacheable tokens and hit rates. Caching is off by default because eligibility depends on your request. Astra requires a cacheable prefix of at least 1,024 visible tokens. OpenAI explains the caching requirements.

Accepted results = attempted tasks × accepted-result rate. At 0%, the calculator still shows spending, but cost per accepted task is undefined. A low token price does not prove that a model completes your workload well.

Where the prices come from

The table above is generated from the same loaded price catalog used by the calculator. It shows each source-check date and expiry, standard rates per million tokens, documented limits and official model-page links. Tiered models list standard rates through 272K input tokens and the published long-context tier above 272K; the GPT-5.4 long-context cached-input price is not stated in the source and cache-hit estimates in that tier are blocked.

KAPI-backed current pricing is paused; this calculator uses the separate starter catalog or your custom rates. Expired or malformed provider prices are excluded. Custom rates are never labeled as provider-verified, and the retry button reloads the local price catalog without changing verification dates.

Inspect KAPI’s August 24 historical snapshot and method. For generation by clip duration, use the AI video cost calculator.