Anthropic Launches Claude Haiku 5.5 with 75% Lower Operating Costs
Anthropic has announced Claude Haiku 5.5, its fastest, most affordable, and most capable lightweight model to date. The new model reduces average operational costs by approximately 75% compared to Haiku 4.5, while Anthropic also halved the prompt cache read pricing for Sonnet 5.5. This drastic cost reduction makes high-throughput LLM workloads and real-time AI applications significantly more economical for developers. Furthermore, Haiku 5.5 is designed to function as a fast sub-agent alongside larger models like Sonnet 5.5 and Opus 5.5, optimizing multi-agent enterprise workflows. For requests containing up to 100,000 tokens, input and output prices are $0.10 and $0.50 per million tokens respectively. It is also the first Haiku model to feature adjustable performance settings, enabling developers to balance cost and intelligence based on their specific needs.
## BACKGROUND
Anthropic categorizes its Claude language models into three main tiers: Opus for maximum capability, Sonnet for balanced performance, and Haiku for high speed and cost efficiency. Modern LLM platforms also utilize prompt caching, which allows developers to reuse pre-processed prompt contexts across API calls to cut processing time and token costs.