Anthropic launches Claude Haiku 5.5: Price slashed by 90%, outperforms GPT-6 Luna in multiple benchmarks.

Beating AI Express: Insights – Anthropic has launched its new lightweight model, Claude Haiku 5.5, focusing on low cost, high speed, and agent tasks. Compared to Haiku 4.5, the new model delivers significant improvements in programming, computer operation, and professional knowledge tasks, with some benchmarks outperforming GPT-6 Luna. Anthropic states this is its fastest and most capable small model to date. Official tests show Haiku 5.5 scored 72.4% in the OSWorld 2.1 computer operation benchmark, higher than GPT-6 Luna’s 48.9% and the previous Haiku 4.5’s 15.7%. In the GDPval-AA professional work benchmark, the new model earned 1620 points, also exceeding GPT-6 Luna’s 1437. Programming capabilities improved notably too: in Terminal-Bench 4.0, it rose from 0% (the prior generation) to 39.2%, surpassing GPT-6 Luna’s 16.4%. Price cuts are steeper. For requests with no more than 100,000 tokens, Haiku 5.5 charges $0.1 per million input tokens and $0.5 per million output tokens, both down 90% from the previous version. Longer requests are priced at $0.5 per million input tokens and $2.5 per million output tokens. Factoring in real-world task token consumption variations, Anthropic estimates average operating costs have dropped roughly 75%. Haiku 5.5 also adds adjustable inference strength for the first time in the Haiku line, letting developers balance performance and cost. The model is now available via Claude API, AWS, Google Cloud, and Azure, suited for batch summarization, information retrieval, browser operations, and executing subtasks for large coding agents.