Gemini 4 Carbon rolls out. Here’s who can use it and what the API costs

Imagine a shop employee comparing the spring and summer product guides and asking which return terms changed and on what pages. Gemini 4 Carbon is the model that reads the documents and writes the answer. The employee can upload the files in a chat, or the shop’s software can send the same files and instructions through an API. The cards show what each route requires to select Carbon.

A token is a unit the model processes when reading text or other material. It does not map one to one to visible characters. For the return-policy comparison, reading the guides contributes to input, and the thinking tokens used to reach the answer are billed at the output rate. Google says Carbon may spend more tokens than Argon on complex work as it runs additional checks.

%% API billing flow · hypothetical guide comparison
flowchart TD
  A["Software · Sends both guides"] --> B["Input · Reads files and request"]
  B --> C["Output rate · Thinking and answer"]
  C --> D["Bill · Input cost plus output cost"]

This hypothetical calculation holds the job at 100,000 input tokens and 20,000 billable output tokens. It uses Standard token rates and excludes caching, search tools and tax. The counts do not come from running the two guides.