Claude Haiku 5.5 Can Now Run High Volume Work For 90 Percent Less

Anthropic has released Claude Haiku 5.5, a small, fast AI model built for high volume, cost sensitive work. It costs up to 90 percent less to run than its predecessor on short tasks, while matching or beating it on speed, giving any business a cheaper way to automate customer support, data entry and research at scale.

What did Anthropic announce?

On 7 October 2026 Anthropic released Claude Haiku 5.5, which it calls its fastest standard speed model yet. For prompts under 100,000 tokens, which covers about 90 percent of typical Haiku traffic, it costs 0.10 US dollars per million input tokens and 0.50 US dollars per million output tokens, against 1 US dollar and 5 US dollars for the previous Haiku 4.5. It is also the first Haiku model with an adjustable effort setting, so a business can choose to optimise a task for cost or for intelligence.

We're very impressed with Claude Haiku 5.5, particularly its speed.from Aaron Vinh, Staff Software Engineer, Asana (Anthropic, 7 October 2026)

What can it actually do?

• Handles high volume, repetitive work well: summarising documents, classifying support tickets, pulling data from records and answering routine customer questions.
• Runs fast enough for live customer support and browser based tasks; Anthropic calls it the fastest model it has released at standard speed.
• Works as a quick subagent alongside larger models like Claude Opus 5.5 for smaller steps inside a bigger task, instead of using an expensive model for every step.
• Is not meant for complex, multi step agentic coding or reasoning; Anthropic still points businesses to Claude Sonnet 5.5 or Opus 5.5 for that kind of work.

How could your business use it?

A retailer could use it to sort and draft replies to routine WhatsApp or email enquiries, saving a more expensive model for genuinely complicated cases. An accounting firm could run it over client invoices and receipts to classify and extract figures before a human checks the summary. A clinic could use it to triage incoming questions and only escalate the ones that need a real answer. Because it costs a fraction of a full sized model, work that was too expensive to automate before, high volume, repetitive and low risk, now has a cheaper way to get built.

Also today

• AMRO raised Malaysia's 2026 growth forecast to 5.1 percent and 2027's to 4.8 percent, pointing to AI related exports and investment as the main driver (The Star, 5 October 2026).
• AWS opened a public preview of Amazon Bedrock Managed Agents built on OpenAI's own Agents API, letting a business run an AI agent on AWS using that framework (AWS News Blog, 5 October 2026).
• Leaked Google app code reported by Android Authority suggests Gemini Gems will stop accepting new edits from 13 October 2026 and fully retire by 17 November 2026, folded into a feature called Skills; Google has not confirmed this.

What should you try this week?

Before you default to the most powerful, most expensive AI model for every task, look at which of your repetitive jobs, sorting enquiries, summarising documents, drafting first replies, could run on a smaller, cheaper model instead. That kind of practical cost thinking is exactly what we cover hands on in Master Claude AI for Business Productivity, where business owners learn to set up Claude for their own day to day admin and reporting work.