Claude Haiku 5.5 costs 90% less per token than Haiku 4.5

Running a high-volume AI task gets expensive fast, and the bill usually comes from the boring parts. Anthropic's new Claude Haiku 5.5 is its fastest Haiku yet, and the price is the part worth your two minutes: it costs 90% less per token than Haiku 4.5 on tasks under 100K tokens, and Anthropic says about 90% of the requests sent to the older Haiku model already fall inside that range.

What Happened

Anthropic released Claude Haiku 5.5. On any task under 100K tokens, it runs 90% cheaper per token than Haiku 4.5.

Anthropic also says roughly 90% of requests made to the previous Haiku model sit inside that token range, so most of the traffic that model was handling fits the cheaper tier.

The model is built for high-volume, cost-sensitive work. It handles repetitive jobs like summaries and classification, and it pairs with Claude Opus 5.5 and Sonnet 5.5 as a subagent on coding work. It is also fast enough for live customer support and browser use.

Why It Matters

Cheaper inference changes what you can afford to leave running. Work you used to batch overnight — summarizing tickets, tagging incoming content, sorting support mail — now fits an always-on flow.

The pairing matters as much as the price. Keep the heavy model for the hard reasoning and point the volume at Haiku 5.5, and the same setup costs noticeably less. If your setup is already on Haiku 4.5, this is the tier where the numbers actually move.