For the last couple of years, "let's add AI to that workflow" has been followed by a nervous glance at the API bill, like ordering lobster at a restaurant with no prices on the menu. Anthropic just slid a much cheaper menu across the table. The small model got smarter, and the price tag shrank instead of growing — a combination about as common as a sequel that beats the original.

Ninety Percent Off, No Coupon Required

Anthropic released Claude Haiku 5.5 on October 7, the first new version of its smallest, cheapest model family in nearly a year. For requests under 100,000 tokens, it costs $0.10 per million input tokens and $0.50 per million output tokens — a 90% cut that lines up with OpenAI's GPT-6 Luna. Above that threshold, pricing steps up to $0.50 input and $2.50 output. Anthropic estimates real workloads come out roughly 75% cheaper than on Haiku 4.5 once you factor in request sizes and token usage.

It's not just a discount, either. Haiku 5.5 brings a one-million-token context window and, for the first time in the Haiku line, an adjustable effort setting so developers can trade speed for deeper reasoning task by task. On the offline subset of the OSWorld 2.1 computer-use benchmark, it scored 72.4%, up from 15.7% on Haiku 4.5. It's available through Anthropic's platform, AWS, Google Cloud, and Microsoft Azure.

The Unglamorous Jobs Just Got Affordable

Anthropic is aiming this squarely at the repetitive stuff: summarizing documents, classifying incoming information, running database queries, and handling the smaller chores inside multi-model agent systems. Think of it as the reliable line cook, while the big flagship models play head chef on the tricky dishes. One customer reported roughly half the latency of the previous version.

Here's the part most people skip: when the cheap model gets this good, the business case for AI changes shape. Tagging every support ticket, summarizing every form submission, or auto-classifying a whole content library used to be "maybe someday" projects. At these prices, they become line items small enough to approve on a Tuesday. The real question shifts from "can we afford AI?" to "do we know which tasks to route to which model?" — and that's a design problem, not a budget problem.

Cheap, fast, and genuinely capable used to be a "pick two" situation. The menu just got a lot more interesting.

If price was the thing keeping AI-powered tagging, summaries, or search off your website's roadmap, our team can help you figure out where a model like this actually earns its keep.

Source: VentureBeat