Claude Haiku 5.5 Slashes Input Costs to $0.10/M

Why this matters

For teams running high-volume pipelines, summaries, classification, subagent orchestration, this pricing changes the math on what's worth automating. At $0.10 per million input tokens, tasks that didn't clear the cost bar on Haiku 4.5 suddenly do. Anthropic's own framing is telling: 90% of Haiku 4.5 traffic stayed under 100K tokens, which means most real-world usage falls into the cheapest tier rather than the one that gets headlines. That's worth noting before anyone assumes the 90% discount applies universally; the realistic blended savings, by Anthropic's own estimate, land closer to 75%.

The tiered structure, with the jump only kicking in past 272K tokens, rewards architectures that chunk work into smaller calls rather than feeding massive contexts in one shot. For founders building on subagents or compaction-heavy workflows, that's a design incentive worth testing against actual token distributions, not list prices. Availability on the API and Bedrock means this isn't a lab demo. The real question for anyone evaluating it: does Haiku 5.5's quality hold up at this price, or is Anthropic betting volume will cover the gap.