Artificial intelligence (AI) startup Anthropic has stoked the battle to win subscribers by releasing an AI model that is cheaper than before.
Anthropic on the 15th (local time) announced the release of its latest small AI model, Claude Haiku 4.5, and said, "A model that was cutting-edge until recently is now cheaper and faster." Compared with Claude Sonnet 4, an AI model released five months ago, Anthropic said the newly released Claude Haiku 4.5 delivers a similar level of coding performance while the expense is about one-third and the speed is more than twice as fast. It added that Claude Haiku 4.5 also outperforms Sonnet 4 in certain tasks such as computer use.
In the SWE-bench Verified test that evaluates AI models, Haiku 4.5's software engineering performance came in at 73.3%, slightly higher than Sonnet 4 (72.7%) and OpenAI's latest model, GPT-5 (72.8%). Mike Krieger, Anthropic's chief product officer (CPO), said in a CNBC interview that day, "Claude Haiku 4.5 punches far above its weight class."
The model is available to free users and charges the lowest price for paid users. Anthropic has been rolling out a family of large language model products under the Claude series, generally naming the smallest model Haiku, the mid-size model Sonnet, and the largest model Opus. The company also released the latest version of its mid-size AI model, Claude Sonnet 4.5, on the 29th last month. It introduced another new model in just about two weeks after that.
Krieger said that while Claude Sonnet 4.5 remains Anthropic's highest-performing model, the newly released Claude Haiku 4.5 is an ideal tool for users who want faster and more accurate answers.