According to a report by The Information, the AI startup Anthropic plans to launch a brand new hybrid AI model in the coming weeks. This model has the ability to flexibly switch between fast response and deep reasoning modes, providing developers with unprecedented control and flexibility.
The standout feature of this model is its adjustable computational cost. Unlike OpenAI's models with fixed settings, Anthropic's model uses a token-based sliding scale that allows developers to precisely control the consumption of computational resources. When the sliding scale is set to "0," the model behaves similarly to OpenAI's GPT-4o, primarily focusing on fast responses; when the sliding scale is increased, the model enters deep reasoning mode.
In deep reasoning mode, the model performs exceptionally well on certain programming benchmarks, even surpassing OpenAI's GPT-3Mini-high. Particularly when handling complex business codebases, Anthropic's model demonstrates a stronger understanding and reasoning capability.
This design allows developers to find the best balance between speed, performance, and cost based on actual needs. For tasks requiring quick responses, a low computational mode can be selected; for tasks needing complex reasoning, deep reasoning mode can be activated.
Some commentators suggest that Anthropic's hybrid model bears similarities to the mixed intelligence concept speculated for GPT-5, but Anthropic's model offers manual control options, granting developers greater autonomy.
The release of Anthropic's new model will undoubtedly bring a new competitive landscape to the AI field and provide developers with more diverse choices. Its features balancing speed, deep reasoning, and cost control are expected to show strong competitiveness in practical applications. This new model is certainly worth looking forward to!
