Anthropic releases Claude Haiku 5.5 small model and halves Sonnet 5.5 cache read prices

Anthropic's introduction of Claude Haiku 5.5 at a significantly reduced cost, alongside lower cache read prices for Sonnet 5.5, highlights a strategic move to optimize AI model economics. By cutting the running cost of Haiku 5.5 by an average of 75% compared to its predecessor, Anthropic aims to make high-volume, repetitive AI tasks more accessible and affordable for enterprises. This focus on cost efficiency is crucial for driving wider adoption in practical applications like customer support and browser automation.

The competitive positioning of Haiku 5.5 against OpenAI's GPT-6 Luna, with reported performance advantages in agentic tests like OSWorld 2.1 and Terminal-Bench 4.0, suggests Anthropic is directly addressing the need for powerful yet economical small models. This direct comparison, backed by benchmarks, provides a clear basis for organizations to evaluate Haiku 5.5 for tasks requiring both efficiency and computational capability. Asana Inc.'s testing, showing over 30% lower latency and up to 2.5 times faster inference, further substantiates these claims.

Beyond pricing and performance, the adjustable effort setting for Haiku 5.5 offers users granular control over cost versus performance. Additionally, the enhanced safety features and cybersecurity safeguards, which reduce misaligned behavior instances compared to Haiku 4.5 and allow more defensive work than Sonnet 5.5, address critical enterprise concerns. These combined factors indicate a comprehensive effort to make Haiku 5.5 a compelling option for practical, secure, and cost-effective AI deployment.