Anthropic has rolled out its newest compact model, Claude Haiku 5.5, heralded as the swiftest, most proficient, and most cost-effective compact model yet. With a starting price of a mere $0.1 per million input tokens, it marks a significant 90% price reduction from its predecessor, translating to an average cost savings of around 75% for comparable tasks. In a series of benchmark evaluations, encompassing computer operations, knowledge-intensive work, and agent programming, Claude Haiku 5.5 has demonstrated superior performance over GPT-6 Luna. This model debuts an adjustable reasoning intensity capability, accommodates a context window extending up to 1 million tokens, and can generate outputs of up to 128,000 tokens. Remarkably, Anthropic accomplished the comprehensive update of the Claude 5.5 series within a mere 15 days, fine-tuning three distinct models tailored for varied scenarios. Haiku 5.5 is meticulously crafted for high-frequency, precisely defined tasks that entail substantial call volumes and are sensitive to pricing, enabling it to function effectively as a sub-agent within intricate agent workflows and substantially diminishing the cost hurdles for extensive deployments. Moreover, Anthropic has slashed the cache read price for Sonnet 5.5 and extended API credits to subscribers, thereby fostering the execution of a greater number of AI tasks that demand high call volumes.
