The single maintained data table behind our calculators: published per-million-token API prices for the main OpenAI, Anthropic and Google model families.
Three of our tools — the API cost calculator, the token estimator and the subscription vs API comparison — all read from this one table. Keeping it in a single place means there is exactly one thing to update when prices move, and one date for you to check before trusting any figure on the site.
USD per 1,000,000 tokens. Last verified 20 July 2026. These figures are maintained by hand and will drift — verify with the provider before budgeting.
| Model | Input / 1M | Output / 1M |
|---|---|---|
| Anthropic | ||
| Claude Fable 5 | $10.00 | $50.00 |
| Claude Opus 4.8 | $5.00 | $25.00 |
| Claude Sonnet 5 | $3.00 | $15.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 |
| OpenAI | ||
| GPT-5.6 Sol | $5.00 | $30.00 |
| GPT-5.6 Terra | $2.50 | $15.00 |
| GPT-5.6 Luna | $1.00 | $6.00 |
| GPT-5.4 mini | $0.75 | $4.50 |
| GPT-5.4 nano | $0.20 | $1.25 |
| Gemini 3.1 Pro | $2.00 | $12.00 |
| Gemini 3.5 Flash | $1.50 | $9.00 |
| Gemini 3 Flash | $0.50 | $3.00 |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 |
Sources: OpenAI API pricing · Anthropic (Claude) pricing · Google Gemini API pricing
Prices are quoted per million tokens, and input and output are billed at different rates. Output is consistently the more expensive side — often three to six times the input rate — which is why a workload that reads a lot and writes a little is so much cheaper than one that generates long responses.
Note also that these are list prices for standard synchronous requests. Real bills are frequently lower, because of two discounts most people miss: prompt caching, where a repeated prefix (a long system prompt, a document you keep asking about) is billed at a fraction of the normal input rate, and batch processing, typically around half price for work that can tolerate a delay.
Model pricing is genuinely volatile. Providers cut prices, introduce cheaper tiers, run introductory rates that expire, rename models and retire others — sometimes several times a year. Any AI pricing article without a date on it should be treated as fiction. Ours carries one, and if the date above is more than a month or two old, check the provider directly before you rely on it.
Spotted a price that has moved? Tell us — corrections to this table are the single most useful thing anyone can send us.