Claude Haiku 5.5

What is disclosed

Claude Haiku 5.5 was released on 7 October 2026, the third model of the Claude 5.5 family.¹¹1Anthropic, Introducing Claude Haiku 5.5, 7 October 2026. https://www.anthropic.com/news/claude-haiku-5-5 Anthropic describes it as built for high-volume, latency-sensitive work such as classification and extraction.²²2Anthropic, Claude Haiku 5.5 model page, Claude Platform docs, read 17 October 2026: released 7 October 2026, 1M context, 128K output, 300K on batch with a beta header, text and images in, text out, adaptive thinking, default effort medium, knowledge cutoff June 2026, retirement not sooner than 7 October 2027, tokenizer of Claude 4.7 and later. https://platform.claude.com/docs/en/models/haiku-5-5/overview Nothing on its size or architecture has been published.

The window holds 1M tokens and a response runs to 128K, or 300K on the Message Batches API with a beta header.²² The model takes text and images and answers in text. Its reliable knowledge stops at June 2026.²²

It uses the tokenizer introduced with Claude Opus 4.7. The same text counts as about 30% more tokens than on Haiku 4.5.²²

What is new

Haiku 5.5 is the first Haiku with an adjustable effort setting.¹¹ Thinking is adaptive and on by default, at medium effort.²² Anthropic presents it as its cheapest and fastest small model, and the most capable of them.¹¹

Input costs $0.10 per million tokens and output $0.50 for prompts up to 100,000 tokens. Above that length the rates are $0.50 and $2.50. Cache reads cost $0.01 and $0.05.³³3Anthropic, Claude Haiku 5.5 model page and Claude API pricing, read 17 October 2026: $0.10 and $0.50 per million input and output tokens up to 100,000 tokens, $0.50 and $2.50 above, cache reads $0.01 and $0.05; Haiku 4.5 at $1 and $5; Sonnet 5.5 at $2 and $10; Batch API 50% off. https://platform.claude.com/docs/en/models/haiku-5-5/overview and https://platform.claude.com/docs/en/about-claude/pricing. Arithmetic: 0.10 / 1 = 0.10 and 0.50 / 1 = 0.50, so the new rates are 10% and 50% of the old ones; 2 / 0.10 = 20. Haiku 4.5 costs $1 and $5.³³ For prompts up to 100,000 tokens the new model is 90% cheaper per token, and for longer ones 50%.³³

Anthropic puts the average saving at about 75% once the heavier tokenizer is counted.¹¹

Its safeguards sit between the two neighbours. The cyber safeguards are stricter than those of Haiku 4.5 and looser than those of Sonnet 5.5. The biology safeguards match Sonnet 5.5.¹¹

Using it

The id is claude-haiku-5-5, the same on the Claude API and on Google Cloud. On Amazon Bedrock it is anthropic.claude-haiku-5-5.²² The Batch API takes 50% off.³³

Code written for Haiku 4.5 needs changes. Manual thinking budgets and sampling parameters return errors, and so do prefilled assistant messages.⁴⁴4Anthropic, What's new in Claude Haiku 5.5, Claude Platform docs, read 17 October 2026. https://platform.claude.com/docs/en/models/haiku-5-5/whats-new-haiku-5-5 Earlier thinking blocks stay in context, so long conversations carry more input tokens.⁴⁴ The smallest cacheable prompt falls from 4,096 tokens to 512.⁴⁴

Claude Sonnet 5.5 sits above it, at twenty times the input price of the cheapest tier.³³ Retirement comes no sooner than 7 October 2027.²²