Anthropic Launches Claude Haiku 5.5 with Adjustable Reasoning

Anthropic has launched Claude Haiku 5.5, the latest iteration of its fastest and most affordable model tier. Operating under the model ID claude-haiku-5-5, the new release introduces a massive 1 million-token context window, up from 200,000 tokens, alongside a maximum output limit of 128,000 tokens. Pricing for the model starts at $0.10 per million input tokens and $0.50 per million output tokens at its lowest effort setting. This base rate represents a 90 percent reduction compared to Haiku 4.5, though Anthropic estimates average workload savings of about 75 percent because higher reasoning settings consume more tokens.

This release marks the first time a Haiku model features adjustable reasoning effort, allowing developers to trade cost for intelligence on a per-request basis. While the default setting is medium, developers can dial the effort down to the lowest setting for simple classification tasks, or scale it up for complex coding subagent operations. This single-model flexibility simplifies application architecture, eliminating the need to route borderline tasks between Haiku and Sonnet 5.5. The model is immediately available on the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure.

For practitioners, Haiku 5.5 is optimized for high-volume operations like customer support, data extraction, and browser control. In multi-agent systems, it can run parallel file inspections or tests while larger models like Opus 5.5 or Sonnet 5.5 act as planners. Alongside this release, Anthropic halved its Claude Sonnet 5.5 cache-read pricing to $0.10 per million tokens, a change expected to lower costs by roughly 20 percent for long-running workloads. Sonnet 5.5 itself boasts output speeds over 30 percent faster than before, with its Terminal-Bench score climbing from 10.3 percent to 70.6 percent.