On October 16, the AI company Anthropic released its new small AI model, Claude Haiku4.5, on its official website. This new model offers performance close to the cutting edge at a very low cost, making it an ideal choice for real-time, low-latency tasks such as chat assistants and customer service.
The Claude series models include three sizes: the small Haiku, the medium Sonnet, and the large Opus. Although larger models have an advantage in depth and knowledge scope, they are slower and more expensive. The Haiku series, on the other hand, uses "distillation" technology to build a small AI model that demonstrates capabilities comparable to large older models in functional tasks such as coding.
In terms of application scenarios, Claude Haiku4.5 perfectly combines intelligence with processing speed, making it particularly suitable for real-time chat assistants, online customer service, and auxiliary programming tools that require immediate feedback. According to official data, the coding ability of Haiku4.5 is comparable to that of Sonnet4, but its cost is only one-third of the latter's, and its processing speed is more than twice as fast.
In the authoritative benchmark test SWE-bench Verified, Haiku4.5 achieved a high score of 73.3%, slightly surpassing Sonnet4's 72.7%. In some specific tasks, its performance even exceeded that of Sonnet4. More surprisingly, in some benchmark tests, Haiku4.5's performance has approached that of OpenAI's GPT-5 model. Nevertheless, the official warned that these data may have been filtered, and readers should view them with caution.
In terms of pricing, Haiku4.5 demonstrates significant market competitiveness. For users who subscribe to the Claude website and app, this model is included in the service. For developers, the API call price is set at $1 per million input tokens and $5 per million output tokens. In comparison, the Sonnet4.5 model in the same series is priced at $3/$15, and the Opus4.1 model is as high as $15/$75. Therefore, Haiku4.5 has become a cost-effective alternative to the older Haiku3.5 and Sonnet4 models.
Anthropic designed an innovative multi-model collaborative workflow for Haiku4.5. In this mode, the more powerful Sonnet4.5 is responsible for breaking down complex tasks into multiple sub-plans and scheduling multiple Haiku4.5 instances to run in parallel. This architecture is similar to a project manager leading a team's collaboration, enabling more efficient and lower-cost solutions to complex problems, opening up new possibilities for advanced application scenarios such as AI agent-based coding.
