LIST TARIFF List prices and the arithmetic behind the bill
Rates verified 18 November 2026 3 calculators 22 models USD list prices Source: vendor and agency pages only

Reference — November 2026

The same thousand calls cost $0.54 or $43.20.

Per-token API prices, platform fees, and US self-employment tax — each one a published rate, the arithmetic between that rate and your bill, and a date on every figure.

Basis: published list rates Negotiated terms and taxes: excluded Each figure: linked to its source page
01Index
  1. 01 AI API cost calculator Requests, input and output tokens, cache reads, cache writes, batch. Twenty-two models, sorted by what your workload would bill, with the line-by-line arithmetic shown for whichever one you select. Live
  2. 02 Platform and payment fee calculator Stripe, PayPal and Etsy across eight rails: percentage and fixed fees on one sale, the take rate at small tickets, and the point where the fixed fee sets your floor. Live
  3. 03 US 1099 self-employment tax estimator Schedule SE line by line: 92.35% tax base, 12.4% Social Security to the 2026–2027 wage base, 2.9% Medicare, 0.9% Additional Medicare, and the half deduction. Live
02Notes

Data notes

One published number per note, the arithmetic it changes, and the page it came from. All notes

  1. 18 Nov Gemini 4 Carbon and Argon double on 1 December $2.00 / $10.00 becomes $4.00 / $20.00: a 2,500-request workload goes from $27.00 to $54.00. Pricing
  2. 18 Nov At $5, the fixed fee is the whole argument Stripe 2.9% + 30¢ against PayPal 2.99% + 49¢, and the ticket below which a fixed fee passes 5% of the sale. Fees
  3. 18 Nov The 2027 wage base adds $967 to a large Schedule SE bill $184,500 to $192,300: the profit level where it starts to matter, and the ceiling shared with a salary. Tax
  4. 18 Nov A cache pays from the first re-read, not from day one Write at 1.25× or 2×, read at 0.1× — the break-even per million cached tokens. AI cost
03Snapshot

A thousand calls, priced

One call = 2,400 input tokens and 600 output tokens, uncached, standard tier. The spread is almost entirely model choice, not prompt length.

Model Provider Input $/1M Output $/1M 1,000 calls
GPT-6 LunaOpenAI0.100.50$0.54
Claude Haiku 5.5Anthropic0.100.50$0.54
Gemini 3.8 FlashGoogle0.753.75$4.05
Gemini 4 HydrogenGoogle1.507.50$8.10
GPT-6.2 SolOpenAI2.0010.00$10.80
Claude Sonnet 5.5Anthropic2.0010.00$10.80
Gemini 4 CarbonGoogle2.0010.00$10.80
Claude Opus 5.5Anthropic4.0020.00$21.60
GPT-6.2 AstraOpenAI8.0040.00$43.20
Claude Fable 5.2Anthropic8.0040.00$43.20

Uncached. Cache reads and batch rates change the ordering only at the margin — read the calculator for your own mix.

04Calendar

Price changes already scheduled

Stated on vendor pages, not announced later.

EffectiveModelChange
30 Nov 2026GPT-5.6 SolPromotional pricing ends.
1 Dec 2026Gemini 4 Carbon, 4 Argon2.00 / 10.00 → 4.00 / 20.00 per 1M.
1 Jan 2027Gemini 3.8 Flash0.75 / 3.75 → 1.50 / 7.50 per 1M.
1 Mar 2027GPT-6.2 Ultrafast3× standard → 6× standard.
05Standards

How these numbers are produced

  1. Vendor pages only. Every rate is read from the provider's own price page on the date stamped at the top of the site. No aggregator, no reseller, no screenshot.
  2. The table is the source. The calculator reads its numbers from the published table on the same page, so an edited rate and a calculated result cannot drift apart.
  3. Changes are dated, not predicted. When a vendor schedules a price step, the date appears here and the calculator keeps using the price in force today.
  4. Gaps are shown as gaps. Where no vendor publishes a rate — a batch cache read, for example — the estimate is labelled "est." in the result rather than presented as a price.