Michael JohnsonAnthropic introduced Claude Haiku 5.5 on October 7, 2026. Input tokens cost $0.10 per million for prompts of up to 100,000 tokens and $0.50 for longer prompts. The company positions the model for high-volume tasks and AI agent workflows; the cost of a specific request depends on its length and the number of tokens processed.
A model for frequent, focused tasks
In its October 7 announcement, Anthropic lists summarization, context compaction, classification, and database queries. Haiku 5.5 is also offered as a subagent for specific steps in coding workflows, while the documentation lists customer support and browser tasks. The model ID on Claude Platform is claude-haiku-5-5.
Anthropic calls Haiku 5.5 its fastest model when comparing standard speed modes; Opus runs faster in Fast Mode. The first Haiku-family model with adjustable effort lets users choose a setting that prioritizes cost or reasoning quality. The editorial team did not conduct independent tests, so speed and quality assessments remain company claims.
Token pricing
Claude Platform pricing is listed per million tokens. The rate depends on prompt length:
| Item | Up to 100,000 tokens | Over 100,000 | Haiku 4.5 |
|---|---|---|---|
| Input tokens | $0.10 | $0.50 | $1 |
| Output tokens | $0.50 | $2.50 | $5 |
| Cache reads | $0.01 | $0.05 | $0.10 |
| 5-minute cache writes | $0.125 | $0.625 | $1.25 |
Compared with Haiku 4.5, token prices are 90% lower for prompts of up to 100,000 tokens and 50% lower above that threshold. Separately, Anthropic estimates that the average cost of completing a task is about 75% lower. The company’s estimate accounts for the distribution of requests and an updated tokenizer, which uses somewhat more tokens for the same task. So the average estimate does not guarantee the same savings for every request.
Recommendation limits and related changes
For complex agentic coding, Anthropic recommends the larger Sonnet 5.5 or Opus 5.5, and Haiku 5.5 for more focused stages, such as summarization and subagent work. This choice allocates costs by type of work: frequent, short operations can go to the smaller model, while more complex tasks stay with the larger models.
As of October 7, according to the same post, Sonnet 5.5 cache reads fell in price from $0.20 to $0.10 per million tokens. Anthropic says the change translates to an approximately 20% reduction in the cost of most agentic tasks. The company also announced that Haiku 5.5 is available on Claude Platform, AWS, Google Cloud, and Azure; the editorial team did not independently verify availability on each third-party platform.
Anthropic said it would begin rolling out monthly API credits for Max and Team subscribers during the week of October 7. As of the October 8 check, completion of the rollout was not confirmed. For teams, the practical takeaway from the launch is to first compare prompt lengths with the 100,000-token threshold, then test the appropriate model on their own tasks and costs.