Antropic Unveils "Claude 3.7 Sonnet," a "Hybrid AI Inference Model" That Gives Users Control Over Its Thinking Process

Unlike existing AI chatbots, this model is attracting attention in that it is a "hybrid AI inference model." This is because it provides the ability to select "immediate answers" and "answers through deeper thinking" depending on how the user wants them.

American artificial intelligence (AI) startup Antropic has unveiled a new AI model, dubbed "Claude 3.7 Sonnet," that marks a departure from the norm. While most AI chatbots operate in a single mode, this model offers a unique "hybrid AI inference model," allowing users to choose between "immediate answers" and "answers through deeper thinking."

The key feature of this model, as highlighted by TechCrunch and other foreign media outlets, is the user's ability to control the model's thinking process. Simple questions are answered quickly, but if a user desires a more in-depth response, the AI can be prompted to spend more time "thinking."

This seems to be an attempt to overcome the limitations of current AI technology. While existing AI models understand and answer questions in a set manner, Sonet 3.7 can provide more accurate and in-depth answers through a logical thought process. This could notably enhance its utility as an AI tool for assisting with complex coding problems or strategic decision-making.

According to Claude, the newly announced model achieves the highest level of performance in the industry. However, this also comes at a premium cost. The pricing is set at $3 per million input tokens and $15 per output token, which is significantly higher than OpenAI's o3-mini model (input $1.10/output $4.40) or DeepSeek's R1 model (input $0.55/output $2.19).

It's also worth noting that the 'inference function' is only available to paid users. Free users can still access Cloud 3.7 Sonnet, but they will be using a typical version without the deep thinking capability.

Antropic's model showcases a potential direction for the development of AI models. Notably, the ultimate goal is to enable AI to independently control the duration of its thinking process. While users currently adjust the accident time, in the future, AI will likely evolve to analyze the difficulty of questions and determine the optimal accident time.