Claude 3.7 Sonnet Model Introduction In March 2024, Anthropic released the third generation of large language models, including Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. The parameter scale increases successively, and the capabilities increase successively. Three months later, in June 2024,Anthropic releases Claude 3.5 Sonnet model, the capabilities were greatly enhanced, and the other two versions were not updated. Four months later, in October 2024, Anthropic released Claude 3.5 Sonnet New, which was equivalent to the version number not changing, but the capabilities were improved. Claude 3.5 Haiku was also released. Claude 3.7 Sonnet is the industry's first large model to support dual output mode,That is, it supports standard output mode and deep reasoning mode.Anthropic believes that humans have never grown two heads for problems that require deep thinking. Therefore, a large model that supports both fast responses that do not require a thinking process and a large model that supports deep thinking is the right direction. The Claude 3.7 Sonnet deep reasoning mode is called Claude 3.7 Sonnet with 64K extended thinking. Note that the 64K here should be the length of the reasoning. Claude 3.7 Sonnet deep inference mode also has an option that developers can control, namelyThe maximum length of thinking that the model is allowed to useThis setting is specified through the System Prompt, and the official Chat page does not support control options. In terms of mathematical reasoning, Claude 3.7 Sonnet is not outstanding, slightly lower than OpenAI o1, Deepseek-R1, etc.. But the gap is not big. But it is very good in common sense reasoning and programming. Especially in programming, in the programming of real-world scenarios such as SWE-Bench, Claude 3.7 Sonnet is much better than other models without using reasoning mode (there is no data in reasoning mode). Other official comparison data (pictures taken from the official): Claude 3.7 is now available to free users, and API interfaces are available on AWS, Google and other cloud services Currently, the Claude 3.7 Sonnet model is available on the official website, and both free and paid users can use it. However, free users cannot use the deep thinking mode. Paid users can. At the same time, in addition to the Anthropic official website, AWS and Google also provide interface services for the Claude 3.7 Sonnet large model (AWS and Google's own large model services should not be very good). In addition, the API interface price of the Claude 3.7 Sonnet model is the same as the previous version, regardless of whether it is deep reasoning or not. The input is US$3 per 1 million tokens, and the output costs US$15 per 1 million tokens.
