Anthropic Launches Claude Haiku 5.5 for Faster, Lower-Cost AI Workloads

Anthropic has announced Claude Haiku 5.5, a new AI model designed to handle large-scale, time-sensitive workloads at a fraction of the cost of its predecessors. The San Francisco-based AI company posted the announcement from its official AnthropicAI account on 7 October 2026.

The company says Haiku 5.5 costs around 75% less to run on average than Claude Haiku 4.5. That’s a striking claim — and one that, if it holds up in real-world use, could shift how businesses and public bodies approach automated data processing.

What Claude Haiku 5.5 Actually Does

Haiku 5.5 sits at the lighter, faster end of Anthropic’s model range. It’s not built for deep reasoning or complex creative tasks — that’s what Claude Sonnet 5.5 is for. Haiku 5.5 is built for volume: text classification, document summarisation, database queries, information extraction, customer routing, and the kind of repeated automated workflows that need to run thousands of times a day without breaking the bank.

The model supports what Anthropic calls an “effort parameter” — a way of adjusting how much thinking the model applies to a given task. That’s useful for businesses that want to dial up accuracy for complex queries and dial it back for simpler ones, keeping costs in check.

It carries a 1-million-token context window, meaning it can process very long documents or large batches of text in a single pass. Maximum output sits at 128,000 tokens.

The Price

Anthropic publishes its pricing in US dollars per million tokens. For prompts up to 100,000 tokens, input costs around $0.10 per million tokens and output costs around $0.50 per million tokens — roughly 8p and 40p in sterling, though the exact GBP figure will shift with exchange rates and applicable taxes.

For longer prompts — those above 100,000 tokens — input rises to $0.50 per million tokens and output to $2.50 per million tokens, or around 40p and £2 respectively.

Anthropic’s 75% average cost reduction figure is the company’s own estimate. Real-world savings will depend on usage volumes, prompt length, whether organisations use caching, and the cost of integration and monitoring on top.

Where It’s Available

Claude Haiku 5.5 is available now through the Claude API. Developers and businesses can also access it through Amazon Bedrock, Google Cloud, and Microsoft Foundry — the three major cloud platforms where Anthropic has distribution agreements.

That broad availability matters. It means organisations already running workloads on AWS, Google Cloud, or Microsoft’s infrastructure don’t need to move platforms to try the model.

The Wider Picture

Anthropic is one of a handful of AI companies — alongside OpenAI, Google DeepMind, and Meta — racing to push down the cost of running AI at scale. The release of cheaper, faster models like Haiku 5.5 is part of a wider pattern: as the big labs compete for enterprise contracts, token prices keep falling and context windows keep growing.

But lower costs don’t automatically mean better outcomes. Civil liberties groups and public-sector watchdogs have consistently raised concerns about AI being used in classification and routing decisions — the exact tasks Haiku 5.5 is marketed for — without adequate human oversight. Errors in automated classification can affect real people: a wrongly routed benefits claim, a miscategorised patient record, a customer complaint that never reaches a human.

Any organisation deploying the model for tasks involving personal data would need to work through data protection law, information governance requirements, and procurement rules before going live.

Anthropic has not published independent accuracy benchmarks for Haiku 5.5 at the time of writing, and the company’s performance claims have not been independently verified.

What This Means for Kent Residents

No Kent-specific deployment of Claude Haiku 5.5 has been confirmed. But organisations such as Kent County Council, Medway Council, and NHS Kent and Medway ICB are exactly the kind of large public bodies that evaluate tools like this for document processing, customer-service routing, and administrative automation — and sharply lower running costs may make the business case easier to argue. Any Kent resident whose data is processed by an AI system has a right to know how decisions affecting them are made, and public bodies would need to demonstrate proper human oversight before any such system goes live.