Anthropic has launched Claude Haiku 4.5, the latest version of its lightweight AI model, offering faster performance at a reduced cost compared to its larger sibling, Sonnet 4.5. The company claims Haiku 4.5 can operate at more than twice the speed of Sonnet 4 while costing only a third as much.
In benchmark tests, Haiku 4.5 scored 73% on SWE-Bench Verified and 41% on the command-line-focused Terminal-Bench. While slightly below Sonnet 4.5, these results are comparable to previous models such as Sonnet 4, GPT-5, and Gemini 2.5. Tests also indicate strong capabilities in tool use, computer tasks, and visual reasoning.
The new model is immediately available on all free Anthropic plans. Its compact design allows multiple Haiku agents to run in parallel or alongside more advanced models, offering flexibility for developers and businesses. Anthropic sees particular value in applications where speed and low server load are critical.
Mike Krieger, Anthropic’s Chief Product Officer, highlighted the potential for new deployment strategies. “Haiku opens up entirely new possibilities for AI in production environments,” he said. “Sonnet can handle complex planning while Haiku-powered sub-agents execute tasks rapidly, providing a complete toolbox for different job requirements.”
Software development tools are expected to benefit most immediately from Haiku 4.5. Claude Code, Anthropic’s existing development AI, already leverages the model for faster performance, and industry leaders such as Zencoder CEO Andrew Filev say it unlocks new use cases.
Haiku 4.5 follows closely on the heels of Sonnet 4.5, released two weeks ago, and Opus 4.1, which launched two months earlier. The previous Haiku version debuted in October 2024.
