AI Cost CalculatorPlan before you build.
Tools & price references

The math and sources, in the open.

Understand what each calculator includes, what it leaves out and how the price records are checked.

Updated

API calculation

All stored token prices use USD per one million tokens. Select rates for the prompt tier and billing mode before multiplying by request volume. A threshold applies to each request, including cached prompt tokens.

requests × [input × (1 − cache share) × input rate + input × cache share × cache-read rate + output × output rate] / 1,000,000 + other monthly costs

Output should include billed reasoning or thinking tokens. Input should include system messages, history and retrieved material. The tool does not automatically model conversation-history growth, images, audio, video or search/tool calls.

Cache and batch limits

The cache share models reads from an existing eligible cache. Writes, storage and minimum-size requirements vary. Unsupported combined batch/cache rates produce an error instead of an invented discount. For newly written tokens already counted as ordinary input, add only the write premium above the counted input charge, plus full storage charges, in other monthly costs. Adding the full write charge again would double-count input.

Token counting

The token counter bundles gpt-tokenizer 4.1.0 and names the chosen o200k_base or cl100k_base encoding. It counts pasted text locally. It does not claim exact API billing, or an exact tokenizer mapping for the latest GPT families, Claude or Gemini. Text-derived model cost is a planning estimate; measured provider usage is preferable.

The token preview decodes pieces separately, so a Unicode character split across tokens can show a replacement character in the piece display without changing the token count.

Automation and voice comparisons

Zapier tasks, Make credits and n8n executions are different units. Feature suitability is not ranked by price. Missing prices, allowances and mandatory components remain unknown. Currency conversion uses only a rate you supply.

Voice totals distinguish the AI component, carrier minutes and separately charged transfer time. Bundled model and speech charges are not added again. A partial subtotal is not an all-in price. Human handling value is a scenario, not a staffing replacement claim.

Source and freshness policy

Price records were checked on 28 October 2026 against the official sources below. Each record carries a source and date. The API tools warn when the snapshot is more than seven days old; known announced price schedules are applied by date. This is not a live feed or a claim of continuous monitoring.

API rates should be reviewed weekly and on announced changes; business plans should be checked at least monthly and before purchase. These are maintenance requirements, not an already-running monitoring service. Static tables need a rebuild after a rate update.

Corrections

If a rate or condition differs from the provider's current page, use the provider's figure and custom inputs. Report the page, model or plan, source URL and checked date through the operator's website. A price correction does not require sharing prompts or private workflow data.