Anthropic's new model offers faster performance at a lower cost
What's the story
Anthropic has unveiled the latest version of its smallest AI model, Claude Haiku 4.5. The new release promises to deliver performance comparable to Sonnet 4, but at a third of the cost and more than twice the speed. The company says that this new model will be especially useful for free versions of AI products, where it can offer powerful capabilities while reducing server loads.
Performance metrics
Haiku 4.5's performance in benchmarks
In internal tests, Haiku 4.5 achieved a score of 73% on SWE-Bench verified and 41% on Terminal-Bench. While these numbers are lower than Sonnet 4.5, they match those of Sonnet 4, GPT-5, and Gemini 2.5 in both cases. The model also performed similarly well in benchmarks for tool use, computer use, and visual reasoning tasks.
Deployment advantages
Model can be deployed in production environments
The lightweight nature of Haiku 4.5 makes it easy to deploy multiple agents in parallel or alongside a more complex model. This opens up new possibilities for deployment in production environments, according to Anthropic's Chief Product Officer Mike Krieger. He said that "it's opening up entirely new categories of what's possible with AI in production environments." The immediate applications are likely to be seen in software development tools where latency is often a critical factor.
Information
Haiku 4.5 follows Anthropic's recent high-profile launches
Haiku 4.5 comes after a series of major launches from Anthropic, arriving just two weeks after Sonnet 4.5 and two months after Opus 4.1, both celebrated as state-of-the-art at release. The previous Haiku version debuted in October 2024.