Built for throughput
Anthropic released Claude Haiku 5.5 on October 7, 2026. The model focuses on fast, repetitive tasks such as classification, summaries, request routing and customer support.
The company says average operating costs are around 75% below Haiku 4.5, accounting for usage patterns and a revised tokenizer. That is not a guarantee of an identical discount for every individual request.
Pricing depends on context length
For prompts up to 100,000 tokens, API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens. For larger prompts, it rises to $0.50 input and $2.50 output.
A revised tokenizer may produce more tokens for the same content. Teams should therefore compare cost per completed task along with accuracy and latency, rather than looking at the rate card alone.
Pairing smaller and larger models
Anthropic presents Haiku 5.5 as a subagent for focused work alongside Sonnet 5.5 and Opus 5.5. It also cut Sonnet 5.5 cache-read prices at the same time.
A layered architecture may lower costs when routine tasks can be delegated reliably while ambiguous or higher-stakes decisions escalate to more capable models.
Measure before deployment
Haiku 5.5 is available through Claude Platform and cloud partners. Developers still need to validate tool support, safeguards, rate limits and output quality in their own settings.
Agent costs reflect retries, tool calls, context growth and human review. Lower per-token prices matter only when the resulting workflow remains dependable.
Official sources and original visuals
The authors' official images and charts are available through the source links. This article's cover is independent editorial artwork, not an official vendor image.