Anthropic launches Claude Haiku 5.5, cutting average running cost about 75%

Cheap, fast small models are becoming the workhorse layer for agentic and high-volume pipelines, and a 75% cost cut on the successor to Haiku 4.5 pushes that layer closer to commodity pricing. The more telling detail is adjustable reasoning effort: it lets builders trade latency and spend against quality per task, which matters most for sub-agents and browser operation where many calls, not one, drive cost. Whether this compresses margins across the small-model segment, and whether rivals match the pricing, is the open question worth watching.