Claude Haiku 5.5 brings new API pricing and credits for Max and Team

On October 7, Anthropic released Claude Haiku 5.5, a model aimed at quick, repetitive tasks. The company also announced cheaper Sonnet 5.5 cache reads and the rollout of API credits for some Claude subscribers. These affect different parts of a bill and should be calculated separately.

Prompt length determines the pricing tier

The Haiku 5.5 documentation lists $0.10 per million input tokens and $0.50 per million output tokens for prompts containing up to 100,000 input tokens, inclusive. Above that threshold, the rates are $0.50 and $2.50 respectively. The threshold concerns one prompt’s input size, not the account’s monthly consumption.

The API model identifier is claude-haiku-5-5. Check the migration guide before switching: changing the model name alone does not guarantee that every parameter from an older request remains compatible.

For your own estimate, take a representative set of tasks and count incoming text, generated output and any retries. Classifying short support requests and summarizing a large document can produce very different cost profiles. A cheaper token alone does not reveal the cost of an acceptable result.

What changed for Sonnet

Sonnet 5.5 cache-read pricing falls from $0.20 to $0.10 per million cached input tokens; the API pricing page confirms the new rate. This does not halve every Sonnet rate: savings depend on how much of the workload uses the cache. For background on the model, see our Sonnet 5.5 article.

API credits do not increase chat limits

Anthropic’s terms provide $100 per month for Max 5x and $200 for Max 20x. Team credits depend on subscribed seats and are pooled, with a $500 cap for the whole team. Eligibility requires an active subscription for at least seven days and linking a Console organization; unused credit does not roll over.

This is an API budget, not extra messages in Claude. For an editor who only uses chat, the first question is whether API automation is useful at all. For a developer, it is how many successfully verified tasks the allowance can cover.