Updates

Claude Haiku 5.5 launches with 2 pricing tiers and API credits for Max and Team

Two streams of token blocks and a circular cache path
Conceptual illustration: Two streams of token blocks and a circular cache path.

Anthropic launched Claude Haiku 5.5 on October 7, targeting high-volume tasks sensitive to response time. Alongside the release, the company reduced Claude Sonnet 5.5 cache-read pricing and announced monthly API credits for Max and Team subscribers. All 3 changes affect usage costs, but in different ways: model pricing, the price of a request component and an account benefit are separate things.

Prices change above 100000 tokens

According to the official pricing table, for prompts of up to 100000 tokens, Haiku 5.5 costs 0.10 USD per million input tokens and 0.50 USD per million output tokens. Above that threshold, prices are 0.50 and 2.50 USD respectively. The lower rate therefore cannot be presented as a fixed price for every prompt length, even though the model supports a 1000000-token context window.

Cache pricing also follows the tiers. For a short prompt, a 5-minute write costs 0.125 USD per million tokens, a 1-hour write costs 0.20 and a read costs 0.01. For a long prompt, the rates are 0.625, 1 and 0.05 USD. Batch input and output cost 0.05 and 0.25 in the short tier, or 0.25 and 1.25 in the long tier. These are component prices, rather than a fixed price for an entire task.

Tokens read from cache do not disappear from the input length. Anthropic's documentation distinguishes ordinary input, cache writes and cache reads, combining them in the total input count. Cost estimates must also account for output and any tools used. The company's estimate of average savings when running the model does not promise the same reduction for every workload or request.

Sonnet cache reads cost less

Sonnet 5.5 cache-read pricing has fallen from 0.20 to 0.10 USD per million tokens. The reduction applies to that component, rather than every line in the pricing table. A request reading large amounts of cached information may be affected differently from one dominated by new input and output. Anthropic estimates savings on typical agent tasks, but actual costs depend on the usage mix.

For developers, the new model identifier is claude-haiku-5-5. The company positions it for high-volume work, while larger models remain relevant to complex tasks. Selecting a model solely by its input price can overlook differences in output quality, the number of correction attempts and the number of tokens a task consumes.

Monthly credits have separate conditions

Max 5x includes 100 USD in monthly API credits, while Max 20x includes 200 USD. For Team, the amount is 20 USD per Standard seat and 100 USD per Premium seat, pooled up to 500 USD per month. According to the Claude help centre, a Claude Console organisation must be linked, and the benefit is rolling out gradually. New subscribers must wait 7 days on an eligible plan.

Credits expire at the end of the billing cycle and do not roll over. They cover eligible API usage on Claude Platform, rather than increasing the regular chat allowance or covering use through external cloud providers. Users should therefore check both their account eligibility and the billing route their application uses. The credit announcement does not itself change the subscription price.