Anthropic Launches Claude Haiku 5.5 With 75% Lower Average Costs
Key Takeaways
- •Ha 5.5 input tokens start at $0.10 per million and output tokens at $0.50 per million, down from $1 and $5 respectively for Haiku 4.5.
- •Anthropic says the model costs 90% less for requests up to 100,000 tokens, a category that made up roughly 90% of Haiku 4.5's request volume.
- •On the OSWorld 2.1 offline computer-use benchmark, Haiku 5.5 scored 72.4%, more than four times Haiku 4.5's 15.7%.
- •Reuters reported the launch is Anthropic's third Claude 5.5 model release in about a month, part of its expansion ahead of a planned initial public offering.
- •Anthropic is separately cutting Sonnet 5.5 cache-read prices by 50%, which it expects to lower Sonnet 5.5 costs on most agentic workloads by about 20%.

Anthropic has launched Claude Haiku 5.5, adding a smaller model to its Claude 5.5 lineup that is designed for high-volume business workloads. According to the company, the model runs at roughly 75% lower average cost than Haiku 4.5, while delivering substantial improvements in computer use and reasoning. Reuters characterized the release as Anthropic's third Claude 5.5 model launch in about a month, underscoring the company's rapid expansion ahead of a planned initial public offering. Details are available on Anthropic's official product page: Claude Haiku 5.5.
Pricing Aimed at High-Volume Workloads
Haiku 5.5 is built for repetitive, latency-sensitive tasks such as classification, summarization, database queries, data extraction and customer support. Because such workloads are billed per token, input and output rates largely determine what they cost to run at scale. Anthropic says the model is also designed to operate as a subagent alongside larger Claude models — a setup in which a bigger model delegates discrete steps of a task to a smaller, faster one.
The pricing changes are significant:
- Input tokens start at $0.10 per million, down from $1 for Haiku 4.5.
- Output tokens start at $0.50 per million, down from $5.
- Requests above 100,000 tokens receive a smaller discount.
Anthropic says the model costs 90% less for requests up to 100,000 tokens — a category that accounted for roughly 90% of the requests handled by Haiku 4.5.
Computer-Use Performance Jumps
The largest performance gain appears in computer-use tasks. On the OSWorld 2.1 offline benchmark, Haiku 5.5 scored 72.4% — more than four times Haiku 4.5's 15.7%.
The model also introduces an adjustable effort setting, allowing users to trade speed and cost for additional reasoning. This gives businesses more control over how much computing power each task consumes.
Anthropic is separately cutting the price of Sonnet 5.5 cache reads by 50%, a change the company expects to reduce Sonnet 5.5 costs on most agentic workloads by roughly 20%.
Business Implications
The launch strengthens Anthropic's push into enterprise AI, where companies increasingly need models capable of handling millions of relatively narrow tasks without the expense of premium systems.
Haiku 5.5 is available through Anthropic's platform and major cloud providers, including Amazon Web Services, Google Cloud and Microsoft Azure. Teams running requests beyond the 100,000-token mark will see smaller discounts than the starting rates suggest, a distinction that shapes realized savings on long-context workloads. Reuters reported that the release adds to Anthropic's efforts to broaden its technology portfolio as it prepares for a potential IPO.