Anthropic Launches Claude Haiku 5.5 With 75% Lower Average Cost

Anthropic has launched Claude Haiku 5.5, a new AI model designed to deliver fast responses, strong performance and lower operating costs for developers and businesses.

Released on October 7, 2026, Haiku 5.5 is built for high-volume tasks where speed and affordability matter. These include summarising documents, classifying information, answering customer questions, extracting data and supporting AI agents.

Anthropic says the model costs around 75% less to run on average than Claude Haiku 4.5. It also introduces adjustable effort controls, allowing developers to balance the amount of reasoning used against cost and performance.

The launch reflects a growing shift in the AI industry. Businesses increasingly need models that can handle thousands or millions of requests without making every interaction expensive.

What Is Claude Haiku 5.5?

Claude Haiku 5.5 is the latest small model in Anthropic’s Claude family.

While larger models are often used for complex reasoning and demanding software development, Haiku 5.5 is designed for tasks that need quick, reliable responses at scale.

Potential applications include:

• Customer support assistants
• Document summarisation
• Information extraction
• Text classification
• Database queries
• AI agent workflows
• Browser and computer automation
• Coding assistance for smaller tasks

Anthropic also positions Haiku 5.5 as a useful supporting model for larger Claude systems. For example, a more powerful model could handle a complex project while Haiku performs smaller tasks such as searching documents, gathering information or summarising results.

Read the official Claude Haiku 5.5 announcement.

Claude Haiku 5.5 Pricing: What Developers Need to Know

Pricing is one of the biggest reasons developers may consider the new model.

For API prompts up to 100,000 tokens, Anthropic lists the following prices:

API usage	Claude Haiku 5.5
Input tokens	$0.10 per million
Output tokens	$0.50 per million

For prompts exceeding 100,000 tokens, the prices increase to $0.50 per million input tokens and $2.50 per million output tokens.

For comparison, Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens. That means Haiku 5.5 is 90% cheaper for prompts up to 100,000 tokens and 50% cheaper for longer prompts. Anthropic estimates an average reduction of around 75%, based on its usage assumptions.

These rates matter for companies running high volumes of AI requests. Even a small reduction in the cost of each interaction can add up when an application processes large numbers of customer messages, documents or automated tasks.

Actual costs will depend on prompt length, output volume, caching and how the model is used.

Stronger Computer Use and Reasoning

Haiku 5.5 is not just a cheaper version of its predecessor. Anthropic reports major improvements across several benchmarks.

On OSWorld 2.1, which measures how well an AI system can operate a computer to complete tasks, Haiku 5.5 scored 72.4% on the offline evaluation subset.

It also scored 45.9% on Humanity’s Last Exam without tools, a challenging test of academic knowledge and reasoning. With tools, Anthropic reports a score of 57.4%.

These are vendor-reported benchmark results, so developers should consider the testing setup and their own evaluations when deciding whether the model fits a particular application.

The results nevertheless suggest that Haiku 5.5 could handle more than basic text generation, especially when paired with tools and carefully designed workflows.

Adjustable Effort Controls Give Developers More Flexibility

Another notable addition is adjustable effort control, available for the first time in a Haiku-class model.

This allows developers to choose how much effort the model should spend on a task. Lower effort settings can suit straightforward, high-volume requests, while higher settings may help with tasks that need more careful reasoning.

For example, a customer service system might use a lower effort setting to categorise routine enquiries. A document analysis workflow could allocate more effort when extracting information from complex material.

The benefit is flexibility. Developers can tune the model for the requirements of each task rather than using the same settings for every request.

Why Claude Haiku 5.5 Matters for AI Agents

AI agents often perform many small operations in sequence. They might search for information, inspect documents, use a browser, update a record and then summarise the result.

If every step relies on an expensive model, the total cost can rise quickly.

Haiku 5.5 offers a more affordable option for tasks such as:

• Routing requests to the right tool
• Summarising intermediate results
• Extracting specific details
• Handling routine browser tasks
• Supporting larger models as a subagent

Its combination of speed and lower cost could help businesses run these workflows more frequently. However, it is not intended to replace larger models in every situation. Anthropic notes that Sonnet 5.5 and Opus 5.5 remain stronger choices for more demanding agentic coding tasks.

Claude Haiku 5.5 vs GPT-6 Luna

Anthropic’s published comparison also includes OpenAI’s GPT-6 Luna on several benchmarks.

For example, Anthropic reports a 72.4% score for Haiku 5.5 and 48.9% for GPT-6 Luna on OSWorld 2.1’s offline subset. On Terminal-Bench 4.0, Haiku 5.5 scores 39.2%, compared with 16.4% for GPT-6 Luna.

These figures come from Anthropic’s own published evaluations and should be interpreted in that context. Benchmark results do not guarantee that one model will perform best across every workload.

For developers, the practical comparison should include response speed, reliability, tool-use performance, total cost and the quality of results on their own tasks.

Where Can You Use Claude Haiku 5.5?

Anthropic says Haiku 5.5 is available through Claude, Claude Code and its API platform. Developers can also access it through Amazon Web Services, Google Cloud and Microsoft Foundry.

The model is available to Claude users across supported plans, with API access for applications and agent workflows.

Developers can review current availability and implementation details through the Claude Haiku product page and the Anthropic Newsroom.

Final Thoughts

Claude Haiku 5.5 strengthens Anthropic’s offering for developers who need capable AI at a manageable cost.

Its lower pricing, adjustable effort controls and reported improvements in computer use and reasoning make it a promising option for high-volume applications and AI agent workflows.

The main consideration is choosing the right model for each task. Haiku 5.5 can handle many routine operations efficiently, while more demanding work may still benefit from a larger model.

As AI adoption grows, the ability to balance performance, speed and cost will become increasingly important. With Haiku 5.5, Anthropic is making a clear case for smaller models that can do more without requiring the same budget as larger systems.

For more technology news, AI updates and gadget launches, follow GeekQu.