Price and Token Card (September 2026)

For professionals and team leads using Claude on a subscription or the API who want to understand what drives their usage and cost.

• Bundle, Meter, Multiplier: what you send, how each token is priced, and how your choice of model multiplies it
• A price table for current models covering input, output, cache writes and cache hits, plus other charges such as batch and web search
• The formula for the cost of one task, a worked example on three models, and a blank row for your own task
• The tokenizer trap: why you compare cost per task, never price per token
• A short checklist for reading your usage report

Why long chats cost more

Everything in the conversation is sent again on every turn. A pasted report is not paid once: it is paid on every turn that follows.