October 16th.Anthropic Today, October 16th, the official network published a book announcing the launch of the mini-protest AI Models Claude Haiku 4.5. The government says the model provides a powerful performance near the front line at very low costTargets are ideal options for real-time, low-delayed tasks such as chat assistants and customer service。
Briefly describe the three sizes of the Anthropic Claude series model: Haiku (small), Sonnet (medium) and Opus (large). The larger model is based on a larger neural network, usually containing deeper context knowledge, but operating at a slower and more costly pace。
The Haiku series models, on the other hand, usually build smaller AI models through "distillation" techniques that enable them to perform large, old models in functional tasks such as coding。
In terms of application scenarios, Anthropic indicates that Haiku 4.5 is a perfect combination of high intelligence and amazing processing speeds to meet users ' demand for instant feedback, and is specifically designed for applications that require high response speeds, such as real-time chat assistants, online client service agents and support programming tools。
In terms of performance, Claude Haiku 4.5 coding performance is essentially the same as Sonnet 4, but at only one third of the cost of the latter, the processing rate has increased more than twice。
1AI quotes Boven that in the authoritative benchmarking test of programming capacity SWE-bench Verified, Haiku 4.5 achieved a high score of 73.3%, slightly ahead of 72.7% in Sonet 4。
Based on the internal baseline data provided by Anthropic, the model isThe performance of certain specific missions (e.g. simulated humans using computers for their operation) has also surpassed that of Sonnet 4。
Even more surprising is that, in part of the benchmark tests, Haiku 4.5's performance is already very close to OpenAI's GPT-5 model, although official reminders are that these self-reported data may have been screened and should be viewed with caution。
With regard to pricing, Haiku 4.5 demonstrates great market competitiveness. The model is included in the service for the subscriber to Claude and App. For developers, the API call price is set at $1 per million input token and $5 per million output。
In contrast, the same series of Sonnet 4.5 models were priced at $3/15, while the higher Opus 4.1 was as high as $15/75, making Haiku 4.5 the old Haiku 3.5 and Sonnet 4 direct alternatives。
More importantly, Anthropic also designed innovative multi-model workstreams for Haiku 4.5. Under this model, the stronger Sonet 4.5 model is responsible for disaggregating a complex total task into a clear, multi-step sub-plan, and then dispatching multiple Haiku 4.5 examples of these sub-tasks in parallel. Such a structure, similar to that in which a project manager leads the work of multiple efficient implementers, can address complex issues more efficiently and at lower cost, opening up new possibilities for advanced applications such as AI proxy coding。