Anthropic Cuts Haiku Costs, Adds Adjustable Effort Controls

ByMason Reed

October 8, 2026

Anthropic’s Haiku 5.5 targets high-volume AI work with adjustable reasoning effort, lower pricing and new browser-use tools, while the company also cuts Sonnet cache costs.

Anthropic has released Claude Haiku 5.5, a small model aimed at repetitive, high-volume tasks where speed and operating cost matter. Its new adjustable effort setting lets developers trade off cost against intelligence, and the model is available through Anthropic’s platform as well as Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic describes Haiku 5.5 as its fastest model to date and says it costs about 75% less to run on average than Haiku 4.5. The savings vary by prompt length: the company’s published pricing is $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, rising to $0.50 and $2.50, respectively, above that threshold. Anthropic says 90% of requests to the previous Haiku model were within the lower-priced range. It also cautions that Haiku 5.5’s updated tokenizer uses slightly more tokens per task, a factor included in its average-cost calculation.

The launch puts pricing controls alongside capability claims. Anthropic reported scores of 72.4% on the offline subset of OSWorld 2.1, 39.2% on Terminal-Bench 4.0, 46.4% on FrontierCode 1.1 Main and 45.9% on Humanity’s Last Exam without tools. Those figures are company-reported benchmarks, not independent evidence that the model will perform equally well across customers’ workloads.

Anthropic positions Haiku 5.5 for summaries, database queries, classification, customer support and browser tasks, as well as subagent work alongside its larger Sonnet 5.5 and Opus 5.5 models. The company says the larger models remain better suited to complex agentic coding. That distinction matters for developers weighing whether a cheaper model can handle a specific step in an application, rather than assuming one model can replace every part of a workflow.

The model is available under the identifier `claude-haiku-5-5`. Anthropic is also adding beta computer-use and browser-use support to its Python and TypeScript SDKs, opening a route for developers to test systems that interact with live interfaces. Such capabilities can make routine workflows faster, but they also place greater importance on how applications authorize actions and limit access to sensitive data. The announcement does not provide a detailed account of how individual developers should manage those risks.

Anthropic paired the release with a cut to Claude Sonnet 5.5 cache-read prices, which it says are down 50%, from $0.20 to $0.10 per million tokens. The company estimates that the change makes Sonnet 5.5 about 20% cheaper on most agentic tasks, where cached context can represent a substantial share of token use.

It is also rolling out monthly Claude Platform API credits for subscribers: $100 for Max 5x users, $200 for Max 20x users and up to $500 pooled among Team users. The credits can be spent on any Anthropic model. Max and Team customers must link a Claude Console organization through billing settings, and the rollout is scheduled to take several days, according to Anthropic’s Help Center.

The announcement is a product and pricing move, not a disclosed funding round. Its significance will depend on whether lower costs and adjustable effort make smaller models practical for more bounded jobs without compromising reliability. For businesses already building on cloud platforms, availability across AWS, Google Cloud and Azure reduces the need to shift infrastructure simply to try Haiku 5.5; it does not, by itself, settle questions of data governance or vendor dependence.

Leave a Reply

Your email address will not be published. Required fields are marked *