Claude Haiku 5.5 costs about 90% less than Haiku 4.5

Anthropic's Claude Haiku 5.5 is priced at $0.10 per million input tokens and $0.50 per million output tokens for prompts under 100,000 tokens of context, about 90% cheaper than Haiku 4.5.

It scores 39.2% on Terminal-Bench 4.0 at maximum effort, against 16.4% for GPT-6 Luna, a strong showing for a small, cheap model on agent-style terminal tasks. Anthropic also reportedly offers a fast mode for Opus 5.5 at about 2.5 times the speed, and Claude Code now compresses the conversation on every turn to cut latency.

The release got little attention next to bigger launches this week, but price cuts of this size change which jobs are worth automating at all.

For teams running agents at scale, small models do most of the work: sorting, extracting, routing and checking. Cutting the price of that layer by an order of magnitude changes the math on a whole pipeline, not just one call.

Our take

A 90% price cut on a capable small model matters more to most budgets than any frontier launch. Cheap and good enough is where most agent steps live.