Anthropic has released Claude Haiku 5.5, the third model in its 5.5 family and the smallest, fastest and cheapest of the line-up. The company prices it at $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100,000 input tokens, and says it costs roughly 75 percent less to run than Claude Haiku 4.5 while posting large gains on computer-use and command-line benchmarks.
The headline capability is 'effort controls'. Haiku 5.5 is the first model in the Haiku tier to offer the setting, which replaces the fixed thinking budget developers used to tune with budget_tokens: rather than allocating a pool of reasoning tokens, a caller picks an effort level and lets the model decide how much thinking a task warrants. Anthropic is pitching the combination of cheap, fast and tunable at agentic workloads — summarisation, classification and other high-volume, cost-sensitive jobs where an agent loops through many small steps.
Availability was broad from day one: the model shipped simultaneously in Amazon Bedrock, Anthropic's own API and third-party coding tools such as Cursor, which reports 48.4 percent on its CursorBench at maximum effort. The launch continues a rapid release cadence at Anthropic — Haiku 5.5 arrives alongside the larger Opus 5.5 and Sonnet 5.5 — and follows a broader market shift toward small, cheap models tuned for driving agents rather than answering single questions.
