The Claude 3.7 -Sonett from Anthropic ends in Openaai and Deepseek in the next big battle of AI

Anthropic I just fired a warning shot OpenaiPresent Deepseek and the entire AI industry with the start of Claude 3.7 sonnetA model that gives users an unprecedented control over how much time a AI spends “Think”Before creating an answer. The publication next to the debut of Claude codeA AI coding agent for command lines signals the aggressive pressing of Anthropic into the corporate AI market-a push, which could change the creation of software and the automation of work.

The missions could not be higher. Last month Deepseek fascinated the Tech world with a AI Fraction of the costsSend by Nvidia Stock fell by 17% and to expose alarms to America’s AI leadership. Now anthropically betting is that precise control over AI argumentation – not only raw speed or cost savings – gives it an advantage.

“We only believe that argument is a central part and a core component of a AI and not a matter of their own that you have to pay separately for access,” said Dianne Penn, who heads the product management for research at Anthropic. “Just like people, AI should deal with both quick reactions and complex thinking. For a simple question how “how late is it?” Should answer immediately. In the case of complex tasks as to the planning of a two-week trip to Italy and the fulfillment of gluten-free nutritional needs-however, it is comprehensible. “

“We see argument, planning and self -correction as separate skills,” she added. “So this is essentially our way of expressing this philosophical difference … Ideally, the model itself should recognize when a problem requires more intensive thinking and adaptation instead of obliging users to explicitly select different argumentation modes.”

The benchmark data support the ambitious vision of Anthropic. In extended mode of thought, Claude 3.7 achieves Sonnet 78.2% accuracy For final tasks that challenge the latest models from Openai and exceed Deepseek-R1.

However, the more revealing metrics come from real applications. The model rated 81.2% when using tools oriented in retail and shows significant improvements in Instructions (93.2%) – Areas in which competitors either have to struggle or have not published any results.

While Deepseek and Openai lead in Traditional Mathematics BenchmarksThe uniform approach of Claude 3.7 shows that a single model can effectively switch between fast reactions and deep analyzes, which means that the need for companies, separate AI systems for different types of tasks can be eliminated.

How Anthropics Hybrid -Ki Enterprise COMPUTING could redesign

The time of release is crucial. Deepseek’s origin sent last month Shock waves Through the Silicon Valley and show that demanding AI far less computing power than thought before. This questioned basic assumptions about the AI ​​development costs and the infrastructure requirements. When Deepseek published its results, Nvidia shares 17% dropped In one day, investors suddenly wonder whether expensive chips were really essential for the advanced AI.

The missions could not be higher for companies. Are companies Spend millions The integration of AI into their operations, bets on which approach the approach dominates. Anthropic’s hybrid model offers a convincing middle path: the ability to optimize the AI ​​performance based on the respective task of immediate customer streaming to complex financial analyzes. The system stops anthropics previous pricing Of $ 3 per million input token and $ 15 per million output token, also with additional argumentation functions.

“Our customers try to achieve results for their customers,” said Michael Gerstenhaber, head of the Anthropic platform. “If you use the same model and challenge the same model in different ways Thompson Reuters In order to carry out legal research, it enables our coding partners such as cursor or Girub Develop applications and achieve these goals. “

Anthropic’s hybrid approach is both a technical development and a strategic gambit. While Openai maintains separate models for different functions and deepseek focuses on Cost efficiencyAnthropic pursues uniform systems that can take on both routine tasks and complex arguments. It is a philosophy that could re -circumvent how companies use AI and eliminate the need to juggle several specialized models.

Meet Claude Code: AIS new developer assistant

Also unveiled anthropically today Claude codeA command line tool with which developers can delegate complex technical tasks directly to AI. The system requires the consent of humans before they commit code changes and reflects the growing industry that focus on responsible AI development.

“You actually have to accept the changes that Claude makes. They are a reviewer with manual exchanges (the) wheel, ”Penn noticed. “There are essentially a kind of checklist that you have to accept essentially for the model to take certain actions.”

The announcements come in an intensive competition in AI development. Stanford researcher recently created Open source argumentation model for less than 50 US dollars, while Microsoft is just integrated Openais O3-Mini model in Azure. The success of Deepseek has also stimulated new approaches to AI development. Some companies examine model distillation techniques that could further reduce costs.

From Pokémon to Enterprise: Testing AI’s new intelligence

Penn illustrated the dramatic progress in AI functions with an unexpected example: “We asked different versions of Claude to play Pokémon. This version has made it so far Vermilion CityCarried several Pokémon and even dragged on level-up. It has the right Pokémon to fight rivals. “

“I think we will see that we continue to innovate and urge the quality of thinking and urge dynamic argument,” said Penn. “We always made it as a central part of intelligence and not as something separate.”

The actual test of the anthropic approach takes place from the introduction of companies. While playing Pokémon Trivial may appear, it shows the kind of adaptive intelligence that companies need: AI that can handle both routine operations and complex strategic decisions without switching between special models. Earlier versions of Claude could not navigate beyond the starting city of a game. The latest version creates strategies, manages resources and hits tactical decisions that reflect the complexity of real business challenges.

For corporate customers, this could mean the difference between maintaining several AI systems for different tasks and the provision of a single, more capable solution. In the next few months it will be shown whether Anthropics bet on uniform AI argumentation will redesign the company market or will only become another experiment in the quick development of the industry.