Anthropic launches Claude Haiku 5.5

What is Claude Haiku 5.5?

Anthropic says Claude Haiku 5.5 is its newest small model and its fastest, cheapest and most capable one so far. The company announced the model on October 7, 2026, positioning it for high-volume work where speed and cost matter more than maximum model size.

According to Anthropic, the model is designed for summaries, compactions, database queries and classification requests. The company also said Haiku 5.5 is its first Haiku-class model with an adjustable effort setting, letting users tune for cost or intelligence depending on the task.

How much does it cost?

Anthropic says Claude Haiku 5.5 costs around 75% less to run on average than Haiku 4.5, and 90% less for requests up to 100,000 tokens. For requests over 100,000 tokens, Anthropic says the model is priced 50% lower than Haiku 4.5.

Anthropic’s platform documentation lists pricing at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 per million input tokens and $2.50 per million output tokens for longer prompts. The company’s pricing page also says batch processing is cheaper, at $0.05 per million input tokens and $0.25 per million output tokens for prompts up to 100,000 tokens.

The company said prompts up to 100,000 tokens account for around 90% of requests to its previous Haiku model, which helps explain why it emphasized lower prices in that range. Anthropic also said Haiku 5.5 uses a newer tokenizer than Haiku 4.5, so the same text counts as about 30% more tokens than before.

Where can developers use it?

Anthropic said Claude Haiku 5.5 is available now on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. Its platform docs list model IDs across those environments, including claude-haiku-5-5 on Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS, plus anthropic.claude-haiku-5-5 on Amazon Bedrock.

The model has a 1 million token context window and can produce up to 128,000 output tokens, according to Anthropic’s docs. AWS’s model card also says the model is offered as a third-party model through AWS Marketplace, with charges appearing on the AWS bill and in AWS Cost Explorer under the model provider.

AWS further lists global and region-specific inference IDs for Bedrock, including global.anthropic.claude-haiku-5-5 and us, eu, au and jp variants. Its model card says Haiku 5.5 also supports compaction on the bedrock-runtime endpoint under the beta value compact-2026-09-04.

How does it compare with Haiku 4.5?

Anthropic says the new model materially outperforms Haiku 4.5 on several internal and benchmark tests while also improving alignment. The company reported 72.4% on the OSWorld 2.1 offline subset versus 15.7% for Haiku 4.5, 39.2% on Terminal-Bench 4.0 versus 0.0%, and 45.9% on Humanity’s Last Exam without tools versus 10.2% for the older model.

With tools, Anthropic said Haiku 5.5 scored 57.4% on Humanity’s Last Exam compared with 18.7% for Haiku 4.5. It also reported 46.4% on FrontierCode 1.1 (Main).

On safety, Anthropic said Haiku 5.5 showed major alignment improvements relative to Haiku 4.5, with fewer instances of misaligned behavior and lower willingness to cooperate with misuse. The company said its cybersecurity safeguards are more restrictive than Haiku 4.5’s but less restrictive than those on some newer models, allowing more defensive tasks than Sonnet 5.5 while still blocking penetration testing. For biology, Anthropic said the safeguards match Sonnet 5, Sonnet 5.5 and Opus 5, allowing research biology questions while restricting harmful requests.

Anthropic also said it is cutting Claude Sonnet 5.5 cache-read pricing in half to $0.10 per million tokens from $0.20, which it said reduces Sonnet 5.5 cost on most agentic tasks by around 20%. The company separately said it will roll out monthly API credits this week to Max and Team subscribers, $100 for Max 5x, $200 for Max 20x and up to $500 pooled for Team users, and that those credits can be used on any of its models to help people build tools, apps and agents that call its API.

Anthropic also said it is updating its Claude Python and TypeScript SDKs with beta support for computer use and browser use.