What is Claude Haiku 5.5?
Claude Haiku 5.5 is Anthropic's Claude model built for high-volume, latency-sensitive API work such as classification, extraction, routing and subagent tasks. Use it to sort support tickets or pull fields from documents at scale. It has a 1M-token context, but long prompts cost more and it counts text as about 30% more tokens than Haiku 4.5.
What can you do with it?
Anthropic positions Haiku 5.5 for work where many small requests must be answered quickly and cheaply. A typical use is a classifier: send each incoming support ticket with a fixed list of labels and get back one label and a confidence note, then route the ticket automatically.
Another is extraction: give it an invoice or contract and ask for named fields as JSON. In agent systems, a larger model such as Claude Opus 5.5 can plan the work while Haiku 5.5 handles repetitive subagent steps. The 1M-token context and 128K output also allow long inputs, though prompts above 100,000 tokens move to a higher price.
Try it on a small task
Suggested workflow: take 50 real tickets or emails you have already labelled by hand, write a prompt with your label list and send them to claude-haiku-5-5 in the Claude Console or API. Leave effort at a low setting first.
Compare its labels with yours and count the mismatches, then check the token counts and cost the API reports. If accuracy is too low, raise effort or add examples before switching to a larger model, and compare cost per correct label rather than price per token.
Pricing and access
Haiku 5.5 is available through the Claude API as claude-haiku-5-5, and Anthropic also lists it on Amazon Bedrock, Claude Models on Google Cloud and Microsoft Foundry. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens; above 100,000 tokens, $0.50 and $2.50. Cache hits cost $0.01 per million tokens for short prompts, and cache writes $0.125 for a 5-minute and $0.20 for a 1-hour lifetime.
If you are moving from Haiku 4.5, Anthropic lists breaking changes: manual extended thinking budgets and assistant message prefill return errors, temperature, top_p and top_k can return errors, and responses can begin with thinking blocks, so select content by type. Safety classifiers can also decline a request with a refusal stop reason, which your client should handle.
What stands out
- 1M-token context window and up to 128K output tokens
- Adaptive thinking with an effort setting
- Prompt caching from 512 tokens
- Browser use tool on the Claude API and Google Cloud
Before you choose
- Prompts over 100,000 tokens are billed at five times the base rate.
- Sampling parameters, prefill and manual thinking budgets return errors.
- Recount tokens: the same text uses about 30% more than on Haiku 4.5.
Plans & pricing
Current pricing on platform.claude.com (opens in a new tab)Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further