Anthropic is launching Claude Haiku 4.5, the newest version of the generative AI ​(genAI) ​company’s smallest language model. The company says it offers the same performance as Sonnet 4, but at a third of the cost and more than twice the speed.

Haiku 4.5 reportedly performs on par with Sonnet 4, OpenAI’s GPT-5, and Gemini 2.5 in several benchmark tests, including SWE-Bench, Terminal-Bench, and tasks related to tool, computing, and visual reasoning skills.

The model is available in all of Anthropic’s free plans. Due to its low resource consumption, multiple Haiku agents can be run in parallel or in combination with more advanced models.

The previous version of Haiku was released a year ago. Anthropic two weeks ago launched Sonnet 4.5 and, two months ago, Opus 4.1.