Another week, and there is another new model of artificial intelligence, ready for public use. This time it is an anthropic with the introduction Claude 3.7 sonnetThe company describes its last release by the first “model of hybrid reasoning” of the market, which means that the new version of Claude can answer the question almost instantly or not rush to go through it step by step. As a user, you can decide which approach the Claude uses, with an opening menu that allows you to choose a “mode of thinking” that you want him to accept.
“We have developed Sonnet Claude 3.7 with another philosophy of other models of reasoning in the market. In the same way, people use one brain both for a quick reaction and for deep thought, we believe that reasoning should be an integrated ability of border models, and not a separate model, ”writes Antrop. “This unified approach also creates a more smooth experience for users.”
Anthropic does not call Openai clearly, but the company clearly shoots the rival. Between GPT-4, O1, O1-Mini And now O3-MiniOpenai offers many different models, but if you do not follow the company, the number of proposed systems can be stunning; In fact, Sam Altman recently admitted the same amountThe field “We hate a model collector as strongly as you, and want to return to Magic Unified Intelligence” – he Published on X Earlier this month.
Anthropic says that it also used another approach to the development of Claude's arguments. “We optimized somewhat less for mathematics and computer science problems, and instead have shifted the focus on real tasks that better reflect how enterprises actually use LLMS,” the company writes. By this moment, current Claude users can count on “especially strong improvements in coding and interface web development.”
Claude 3.7 Sonnet is available for use, starting from today in all Claude plans, including the free level of Anpropic. Meanwhile, developers can access the new model through the company's API, Amazon Bedrock and Google Cloud Vertex AI.
Speaking about the developers, Anpropic is also introduced by Claude Code, a new “agent” tool that allows you to delegate coding tasks in the Claude directly from the terminal interface. Currently, accessible as a limited preliminary viewing of research, Claude Code can read the code, edit files, write and start tests and even push to Github.