Updated
Updated · KDnuggets · Oct 19
Anthropic Launches Claude Haiku 5.5 With 90% Price Cut and 1M-Token Context
Updated
Updated · KDnuggets · Oct 19

Anthropic Launches Claude Haiku 5.5 With 90% Price Cut and 1M-Token Context

3 articles · Updated · KDnuggets · Oct 19

Summary

  • Claude Haiku 5.5, released Oct. 7, is priced at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, which Anthropic says makes it its cheapest and fastest small model.
  • The upgrade expands context to 1 million tokens from 200,000, doubles max output to 128,000, and adds adaptive thinking with an effort setting; Anthropic also reported large gains on its internal benchmarks versus Haiku 4.5.
  • Migration can break existing code: temperature, top_p, top_k, assistant prefills, and manual thinking budgets now error out, while replies may begin with thinking blocks and refusals must be handled explicitly.
  • Anthropic is pitching the model for high-volume tasks such as classification, extraction, summaries, database queries, and sub-agent workflows, with availability across its own platform, AWS, Google Cloud, and Azure.
  • Actual savings may be closer to 75% on average because the newer tokenizer counts roughly 30% more tokens, and larger Claude models still outperform Haiku 5.5 on more complex coding and agentic work.

Insights

Could the hidden pricing surge for massive context windows secretly negate the advertised cost savings of Haiku 5.5?
Why did Anthropic suddenly strip away standard developer controls like temperature in their most capable lightweight AI?
How did a budget-friendly AI model designed for simple tasks manage to exploit software flaws during testing?