Anthropic has released Claude Haiku 5.5, a new lightweight model positioned around low cost, high speed and agent tasks. The company described it as its fastest and most capable small model to date.
Compared with Haiku 4.5, the new model shows large gains in coding, computer-use and professional knowledge workloads. In some benchmarks, it scored above GPT-6 Luna.
Benchmark results moved sharply higher
According to Anthropic’s published tests, Claude Haiku 5.5 reached 72.4% on the OSWorld 2.1 computer-use benchmark. That was above GPT-6 Luna at 48.9%, while the previous-generation Haiku 4.5 posted 15.7%.
On the GDPval-AA professional work benchmark, Claude Haiku 5.5 scored 1620, again above GPT-6 Luna’s 1437.
Coding performance also improved. In Terminal-Bench 4.0, Claude Haiku 5.5 rose from 0% in the previous generation to 39.2%, beating GPT-6 Luna’s 16.4%.
Pricing was cut by 90%
Anthropic also announced a much lower price point. For requests with input lengths of up to 100,000 tokens, Claude Haiku 5.5 costs $0.1 per million input tokens and $0.5 per million output tokens. Anthropic said both figures are 90% lower than the previous generation.
For requests above that length, pricing is set at $0.5 per million input tokens and $2.5 per million output tokens.
After factoring in token consumption in real-world tasks, Anthropic estimated that average operating costs fall by about 75%.
Adjustable reasoning added to the Haiku line
Claude Haiku 5.5 is also the first model in the Haiku family to add adjustable reasoning intensity, giving developers a way to trade off between performance and cost.
Available through API and major cloud platforms
The model is now available through the Claude API, AWS, Google Cloud and Azure. Anthropic said it is suited to batch summarization, information retrieval, browser operation and handling subtasks for large coding agents.