Haiku 5.5 is an Anthropic AI model for teams and developers handling high-volume, latency-sensitive work. It is designed for tasks such as classification, routing, extraction, and subagent work, as well as summarization, text generation, support, automation, and focused coding. It addresses the need to apply these capabilities across repeated workflows, with an adjustable effort setting and options for processing longer prompts. It is available through Claude products, the Claude API, and named cloud platforms.
What can Haiku 5.5 do?
Haiku 5.5 supports adaptive thinking controlled with an effort parameter, which defaults to medium. Its context window supports long prompts, and it can produce long outputs. Anthropic identifies classification, routing, extraction, and subagent tasks as intended work, and also describes summarization, text generation, live support, voice agents, browser and desktop automation, and focused coding.
- Classify content, route requests, or extract information in high-volume workflows.
- Summarize material and generate text.
- Support live support and voice-agent applications.
- Automate browser and desktop tasks or assist with focused coding.
- Use prompt caching and batch processing, which Anthropic says can reduce costs.
It is listed as available in Claude apps, Claude Code, the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Anthropic’s release announcement also described beta SDK support for computer and browser use.
Who is Haiku 5.5 best for?
Consider Haiku 5.5 when a product or team needs to run classification, routing, extraction, or other repeated AI tasks, or wants support for summarization, text generation, live support, voice agents, browser and desktop automation, or focused coding. Its named API and cloud availability may fit teams choosing among several deployment options. Anthropic says Sonnet 5.5 and Opus 5.5 are better choices for complex agentic coding tasks, so evaluate those alternatives for that kind of work.
Haiku 5.5 pricing in brief
The verified API pricing model charges separately for input and output, with rates that vary according to prompt length. The supplied facts do not verify a standalone subscription price for using Haiku 5.5 in Claude apps. Prompt caching and batch processing are available; Anthropic describes them as potential ways to reduce costs. Check the applicable API pricing before estimating a workflow’s spend.
Limitations to know about Haiku 5.5
Anthropic says its other named models, Sonnet 5.5 and Opus 5.5, are better choices for complex agentic coding. Cybersecurity safeguards block penetration testing and other techniques more likely to be used by attackers. The system card reports over-refusal compared with other models in its automated behavioral audit and gives a June 2026 knowledge cutoff. Anthropic also notes that its newer tokenizer can use more tokens for the same work than Haiku 4.5, affecting effective cost comparisons.
Alternatives to Haiku 5.5
Readers comparing AI assistants can also review Claude and ChatGPT. These are options to consider alongside Haiku 5.5 when comparing tools for conversational AI tasks. For coding-focused evaluation, Anthropic specifically identifies Sonnet 5.5 and Opus 5.5 as better choices for complex agentic coding tasks; compare the fit of each option against the work you need to do.
