Anthropic's quick, inexpensive 4.5 Claude Haiku

The Claude Haiku 4.5, the smallest model in Anthropic's most recent lineup, offers performance on pace with the company's flagship model from only a few months ago at a substantially lower cost and faster speeds. The specifics: Claude Sonnet 4's coding skills from May are matched by the new model, which costs only $1 per million input tokens instead of $3. On metrics including computer use, math, and agentic tool use, Haiku outperforms Sonnet 4 despite its small size; on several tests, it even approaches GPT-5. The recently released Sonnet 4.5 can be used as a coordinator for complex activities, allowing enterprises to coordinate numerous Haiku agents working in tandem. All Claude tiers, including free users, can access Haiku 4.5 using the company's Claude Code agentic programming tool and API. Haiku continues to follow the trendline of the nirvana of "intelligence too cheap to meter." With a small, inexpensive model now able to perform at levels that demanded premium price only a few months ago, Anthropic's most recent release demonstrates how rapidly the economics of the AI sector are changing.