Cost calculator
Enter tokens per request and how often you call the model. The estimate uses the rate stored on that row and ignores cache hits, batch discounts, and long-context surcharges. Those are described on the model page.
Loading the calculator…
Questions
How does the cost calculator work?
It multiplies the input and output rates stored on the model you pick by the token counts you enter, then multiplies that request cost by the number of requests. The form starts at 2,000 input tokens, 800 output tokens, and 1,000 requests per day. A month is 30 days, so the starting monthly figure is 30,000 requests. Cached-input prices, batch discounts, and long-context surcharges are not applied. If the stored rate is off-peak or introductory, the estimate uses that labelled rate.
How are prices sourced?
Prices, context limits, and release dates come only from official provider pricing pages, model cards, changelogs, or announcements. ModelMeter does not copy another aggregator. Each row stores the source link and the date that source was checked. If a value is not on the official page, it is left empty and shown as n/a. One number rarely describes every rate a provider charges. The figure in the table is the specific rate named on that row — standard, introductory, off-peak, or legacy. The note records the other published rates, such as long context, batch, peak hours, or a scheduled price change. Hosted open models are labelled with the company that bills the API. A Together or Groq price is not a Meta list price. ModelMeter does not sell API access, run benchmarks, or publish reviews. The check date for the current ledger is 27 Oct 2026. The methodology is also written at https://modelmeter.live/about, and every model page links to the official page the number came from.
What are tokens?
API providers meter usage in tokens, and this ledger stores prices in the unit those providers publish: USD per 1 million tokens. ModelMeter does not convert words or characters into tokens. Tokenizers differ by provider, and this dataset does not include a conversion rate. The calculator uses the token counts you type in.
What does 1M tokens cost?
The input and output columns are already the price of 1 million tokens for the rate named on that row. One million input tokens cost the input rate. One million output tokens cost the output rate. A request that sends both is the sum of those two rates, which is what the calculator computes. For a worked example from this ledger, one request of 1,000,000 input tokens and 1,000,000 output tokens of Claude Sonnet 5.5 (Anthropic) is $12.00 at the stored rate (Base input / cache hit / output): $2.00 input plus $10.00 output. Cache, batch, long-context, peak, and prices scheduled after an introductory window are not folded into that sum. When the official page published them, they are in the model note. On the ledger checked 27 Oct 2026, the lowest published input rate is $0.10 per 1M tokens, on GPT-6 Luna (OpenAI, Standard, under 272K, $0.10) and Claude Haiku 5.5 (Anthropic, Prompts up to 100K tokens, $0.10). The highest published input rate is $10.00 per 1M tokens, on GPT-6 Astra (OpenAI, Standard, under 272K, $10.00) and Claude Fable 5.1 (Anthropic, Base input / cache hit / output, $10.00). 2 models have no published input or output rate, so they are shown as n/a and left out of this range.