GET /v1/models lists the same prices in micros, a millionth of a dollar. Morse writes this page from the prices production bills, so it changes when they do.
How chat is priced
Chat prices are per million tokens. Input is the prompt, and cached input is the part of it the model’s vendor reads from its prompt cache. Output counts every token the model writes, its reasoning included. Context is the most tokens a request can hold, prompt and output together. The same model at two privacy tiers has two ids, and a row for each. A “Prompt over” row is the price of a long prompt. Once a prompt passes that length, the whole request, its output included, pays those prices. On most models, the prompt tokens the vendor writes to its cache cost the input price. These models charge more for them, per million tokens.
Before a chat starts, Morse holds the most it could cost: the prompt at the cache-write price, plus the output limit. Set a lower
max_tokens to hold less. When the reply ends, Morse charges what the tokens cost and returns the rest. See Billing.
How images are priced
Each image costs the price per image, plus the price per reference image for each reference image you send. Each image costs at least one credit, and you pay only for the images delivered. The price per image is what Morse holds for each image. Most image vendors report what each image cost, and Morse then charges by that report, so an image can cost a little more or less than its listed price.How web search, page reading, and transcription are priced
Each call costs a flat price, on top of any chat it serves. A failed call costs nothing.Every price as a table
Chat models
Chat models
Image models
Image models
Web search, page reading, and transcription
Web search, page reading, and transcription