Haiku 5.5 targets narrow agent tasks with adjustable effort

Anthropic has released Haiku 5.5 for tasks such as classification, summarization and focused agent lookups. It adds adjustable reasoning effort to the Haiku family, allowing developers to trade more computation for a harder task. The company positions it for high-volume work while acknowledging that its larger models remain stronger choices for complex coding.

Anthropic · Oct 7, 2026

Why it matters

A smaller model can be useful when a workflow contains many bounded steps rather than one difficult open-ended problem.

More context

Try a task-specific evaluation through a supported provider before changing routing. The announcement lists Claude Platform, AWS, Google and Azure access. Claude Code 2.1.293 also adds model support. Check provider pricing and prompt-length tiers against your actual requests.

Keep in mind

Performance comparisons are vendor results, not our tests. Lower token prices do not establish lower total workflow cost if retries or escalations increase. This is a released model, not a claim that every agent task should use it.

Original read