Anthropic Releases Claude Haiku 5.5 With Sharp Price Cuts
Anthropic's new Claude Haiku 5.5 dramatically improves computer-use performance while cutting token prices by up to 90 percent, though a new tokenizer reduces some of the savings.
Anthropic has unveiled Claude Haiku 5.5, a new version of its lightweight AI model that delivers a dramatic leap in computer-use capabilities alongside steep price reductions. The model's performance on the OSWorld computer use benchmark jumped from 15.7 percent in its predecessor to 72.4 percent, a more than fourfold improvement that positions it as a serious contender for automating desktop tasks.
The price cuts are equally striking. Token prices have dropped by up to 90 percent compared with the previous generation, a move that intensifies the ongoing pricing battle among AI providers. However, the company also introduced a new tokenizer that consumes more tokens per task, partially offsetting the headline savings. The net effect is that users will still pay significantly less for most operations, but the actual cost reduction may be smaller than the raw price list suggests.
The OSWorld benchmark measures an AI model's ability to navigate and operate a computer interface, including clicking, typing, and managing windows. Haiku 5.5's score of 72.4 percent represents a substantial improvement over the 15.7 percent achieved by its predecessor, indicating that Anthropic has made major strides in making smaller, cheaper models viable for complex computer-use tasks. This could accelerate the adoption of AI agents that can perform routine office work, customer service, and other digital chores without human intervention.
Anthropic's decision to slash prices so aggressively reflects the fierce competition in the AI industry, where providers are racing to offer the most capable models at the lowest cost. The pricing arms race shows no signs of slowing, as companies seek to attract developers and enterprises building on their platforms. For businesses evaluating AI solutions, the falling costs and rising capabilities mean that tasks once considered too expensive to automate may now become economically feasible.
The new tokenizer, while improving efficiency in some respects, adds a layer of complexity for users trying to estimate costs. Because it consumes more tokens per task, the effective price per completed action may not fall as much as the per-token price. This nuance highlights the importance of benchmarking total cost of ownership rather than focusing solely on sticker prices. Anthropic has not yet detailed exactly how the tokenizer changes token counts across different workloads, but early analysis suggests the impact varies by task type.
Claude Haiku 5.5 is available now, and its release underscores how quickly the AI landscape is evolving. With each new generation, models are becoming more capable and more affordable, bringing advanced automation within reach for a wider range of organizations. The competitive pressure is likely to drive further innovation and price adjustments across the industry, benefiting users who rely on AI for both simple and complex tasks.
5
