Yesterday’s Claude Haiku 5.5 launch gives agent stacks a worker model good enough and cheap enough to call on every turn, which matters more to builders than another frontier leaderboard. Anthropic says it costs around 75% less to run than Haiku 4.5 on average, and it is the first Haiku with adjustable effort, so each task can lean toward cost or intelligence.

In early testing, customers describe it triaging bugs, handling CRM tasks and pulling a segment revenue line from a 10-K while a bigger model builds the deck.

In the agents I deploy with enterprise teams, most turns are small lookups and summaries around a few hard steps, and a default tier cheap enough for all of them leaves the bigger model free for the work that needs it.