Claude Sonnet 5.5

Five-minute cache writes cost $2.50 per million tokens, one-hour writes cost $4, and cache reads cost $0.10. The Batch API discounts input and output by 50 percent, to $1 and $5, and does not replace the cache-write rates. One million cached tokens that are written for 5 minutes and read once cost $2.60, which is more than the uncached $2 input rate.

The Claude API pricing table does not list a second prompt-length tier for Sonnet 5.5. Input stays $2 and output stays $10 per million tokens. Haiku 5.5 is the model in this family with a separate rate above 100,000 prompt tokens. Bedrock and other hosts are listed as their own model IDs and are not given a dollar rate on the Anthropic page.

A 5-minute cache write is $2.50 per million tokens and each read is $0.10, against $2 uncached. One read costs more than not caching. Two full reads of a 5-minute write cost $2.70, which is less than two uncached sends at $4. A 1-hour write is $4, so it needs more than two full reads before it beats sending the tokens again. Batch halves input and output only, not these write rates.