Anthropicic publishes a new frontier AI model called Claude 3.7 Sonnet, which is a company intended for “thinking” about questions as long as users want it.
Anthropically, they call Claude 3.7 sonnet the first “hybrid model of the hybrid AI reasoning”, because it is one model that can give and real -time answers and more considered, “thought out” answers to the questions. Customers can choose whether to activate the “disperse” of AI models, which Claude 3.7 sonnets asks to “think” for a short or long time.
The model represents anthropin wider effort to simplify the user experience over its AI products. Most Ai chatbota today has a terrifying model voter that forces users to choose several different options that differ in costs and abilities. Laborators like anthropy would rather not have to think about it – in ideal, one model does all the job.
Claude 3.7 Sonnet develops all users and developers on Monday, Anthropopi said, but only users who pay for Anthropic’s top -notch Claude Chatbot plans will get access to the model features. Free Claude users will get a standard, unregistered version of Claude 3.7 sonnets, which anthropic claims outweigh their previous border AI model, Claude 3.5 sonnet. (Yes, the company skipped the number.)
Claude 3.7 sonnets costs 3 million input tokens (which means you could bring about 750,000 words, more words than the entire Lord of the Rings series, in Claude for $ 3) and $ 15 on one million output tokens. This makes it more expensive than the Open-Momi-Domi ($ 1,10 per 1m input tokens/$ 4.40 per $ 1 million output tokens) and Deepseek’s R1 (0.55 USD 1m input tokens/2.19 USD 1m output tokens), but keep in mind that O3 -Di and R1 are models strictly reasoning – not hybrids like Claude 3.7 sonnets.
Claude 3.7 sonnet is the first ANTHOPSKI AI model to “distinguish”, technique Many AI laboratories have turned as traditional methods of improvement AI performance cone.
Models of explanations such as O3-Dom, R1, Google’s Gemini 2.0 Flash Thinking, and XAI’s Grok 3 (Think) use more time and a computer power before answering questions. The models break down less steps problems, which tend to improve the accuracy of the final answer. The explanation models do not think or do not think as if a person, necessarily, would be modeled after deduction.
In the end, Anthrops would want Claude to realize how long he should “think” about questions, without the need for users to choose controls in advance, an anthropin product and a product and research leadership said in an interview.
“Similarly, people do not have two separate brains for questions that can be answered immediately in relation to those who require thought,” Anthropic wrote in a blog post divided by Techcrunch, “we consider explanation to one of the options that would be a border model should be smoothly integrated with other options, not something that can be provided in a separate model. “
Anthropopia says Claude 3.7 Sonet to show its inner planning phase through a “visible scratch pad”. Lee said that TechCrunch users would see Claude’s full thought of thinking for most of the instructions, but that some parts can be renovated for trust and security purposes.
Anthropopia says Claude’s mindset has optimized for tasks in the real world, such as severe coding problems or agencies. API API -Antopica can control the “budget” to think, trading speed and the quality of response.
On one test for measuring the task of coding the actual word, the Swe-clip, Claude 3.7 sonnet was 62.3% correct, compared to the Openi O3-Dom model that achieved 49.3%. On the second test for measuring the AI model ability to communicate with simulated users and external API in retail environment, Tau-Bench, Claude 3.7 Sonet achieved 81.2%, compared to Openi’s O1, which achieved 73.5 %.
Anthropic also says that Claude 3.7 sonnet will refuse to answer questions less frequently than its previous models, claiming that the model is capable of making more tinnamed differences between harmful and benign instructions. Anthropa says he reduced unnecessary rejection by 45% compared to Claude 3.5 sonnet. It comes at a time when Some other AI laboratories reconsider their approach to limit your AI chatbot response.
In addition to Claude 3.7 sonnet, Anthropic also releases an agent coding tool called Claude Code. Starting as a research inspection, the tool allows developers to perform certain tasks through Claud directly from their terminal.
In demonstration, anthropic employees have shown that Claude Code can analyze the encoding project with simple commands such as, “Explain this project structure. “Using ordinary English in command line, the developer can change the codes base. Claude Code will describe its changes as it brings changes and even testing the project for mistakes or push it into a Github repository.
Claude Code will initially be available to a limited number of users based on “The First Come First Servings”, spokesman for anthropa told Techcrunch.
Anthropic releases Claude 3.7 sonnet at a time when AI laboratory delivered new AI models of Breakneck Temp. Anthropa has historically taken a more methodical approach to security. But this time, the company wants to run the package.
How long has the question. Openii may be close to release your own hybrid AI model; The company’s executive, Sam Altman, said he would arrive in “months”.