Anthropic’s Claude Haiku 4.5 matches May’s frontier model at fraction of cost (arstechnica.com)

Anthropic has released Claude Haiku 4.5, a smaller LLM that the company says matches the coding capabilities of its mid-tier Sonnet 4 (and even GPT-5 on coding tasks) from five months ago while running at over twice the speed and roughly one-third the cost. Haiku 4.5 is already available to Claude app, web and API users; the performance claims come from Anthropic’s benchmarks and will be meaningful to verify independently. The announcement highlights a practical trade-off: Haiku targets fast, cheap, task-focused inference (especially coding assistance) rather than the deeper contextual knowledge of larger Sonnet or Opus models. Technically, Haiku 4.5 is a product of model distillation — compressing capabilities from larger networks into a smaller, more efficient architecture — which can preserve functional performance on specific tasks while shedding stored world knowledge. For developers and product teams this matters: lower latency and inference cost make high-volume, real-time code generation and assistants cheaper to deploy, but with the caveat that smaller distilled models can be weaker on complex reasoning or broad-domain knowledge. If independent tests confirm Anthropic’s numbers, Haiku 4.5 underscores how distillation and model lifecycle engineering are reshaping trade-offs between capability, cost, and speed in production AI.