Anthropic has released Claude Haiku 5.5, its fastest small model yet, targeting high-volume workloads such as classification, extraction, summaries, database queries, browser use, and coding subagents.
The model brings adaptive effort to the Haiku line alongside a one-million-token context window, 128K maximum output, vision, and tool use. It is available through the Claude API as claude-haiku-5-5, Amazon Bedrock, Google Cloud, and Microsoft Azure.
For prompts up to 100,000 tokens, API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens—90% below Haiku 4.5’s list price. Longer prompts cost $0.50 and $2.50 respectively. Cache reads and writes follow the same fivefold increase beyond the 100,000-token threshold.
Anthropic reports 72.4% on the OSWorld 2.1 offline subset and 39.2% on Terminal-Bench 4.0. These are vendor-reported launch results, so developers should validate the model against their own workloads.
Why it matters
Haiku 5.5 could materially change the economics of agent systems. Builders can reserve larger models for planning and difficult reasoning while assigning repetitive tool calls, retrieval, routing, compaction, and parallel subagent work to a faster, cheaper model.
That creates a practical model hierarchy: Haiku handles the high-volume execution layer while a larger model such as Claude Sonnet 5.5 handles more demanding everyday agent work. The right split depends on each task’s accuracy, latency, and escalation requirements.
The important caveat is the fivefold price increase once prompts exceed 100,000 tokens. Production teams should measure total task cost—including context growth, cache behavior, and output length—not just headline token pricing.