Anthropic I just fired a warning OpenaiIN Deepseek and the entire AI industry with launch Claude 3.7 SonnetA model that gives users with unprecedented control over how much time AI spends “thinking” before generating answers. Edition, next to the debut Claude CodeThe agent encoding the command line, indicates Anthropica’s aggressive pressure to the AI Enterprise-Phanień market, which could transform the way of building software and work automation.
The rate can’t be higher. Last month, Deepseek stunned the world of technology with the AI model, which suited the capabilities of American systems in A fractionsending nvidia decrease by 17% and raising alarm about the leadership of AI America. Now Anthropic assumes that precise control over the reasoning of artificial intelligence – not only harsh speed or cost savings – will give it an advantage.
“We only think that reasoning is the basic part of AI, not a separate thing that you have to pay separately for access,” said Dianne Penn, which conducts product management for research in Anthropic, in an interview with Venturebeat. “Like people, artificial intelligence should cope with both quick answers and complex thinking. To get a simple question, such as “what time is it?” It should answer immediately. But in the case of complex tasks-as such as planning a two-week travel in Italy while satisfying the dietary needs of gluten-free-implements a more large processing time. “
“We do not perceive reasoning, planning and self-extinguishing as separate possibilities,” she added. “Basically, this is our way of expressing this philosophical difference … Ideally, the model itself should recognize when the problem requires more intensive thinking and adaptation, instead of requiring users to clearly choose different reasoning modes.”
Benchmark data confirms the ambitious vision of Anthropic. In the prolonged pondering mode, Claude 3.7 Sonet reaches 78.2% accuracy In the tasks regarding reasoning at the level of graduates, he questions the latest OPENAI models and exceed the Deepseek-R1.
But more revealing indicators come from application in the real world. Model results 81.2% in the use of concentrated tools and shows clear improvements in Maintaining instructions (93.2%) – Areas where competitors have either fought or didn’t publish the results.
While Deepseek and OpenNai lead Traditional mathematical reference studiesUnified Claude 3.7 approach shows that a single model can effectively switch between quick answers and deep evaluation, potentially eliminating the need to keep up separate AI systems for differing kinds of tasks.
How hybrid artificial intelligence of anthropics can transform the calculations of enterprises
The release time is crucial. The appearance of Deepseek sent last month Shock waves through the Silicon Valley, showing which you can achieve sophisticated AI reasoning Much less computing power than previously thought. This questioned the basic assumptions regarding the cost of AI development and infrastructure requirements. When Deepseek published his results, nvidia supplies fell by 17% In one day, investors suddenly ask if expensive systems were really essential for advanced artificial intelligence.
In the case of firms, rates can’t be higher. Companies are Spending tens of millions Integration of artificial intelligence with their operations, betting which approach will dominate. The Anthropic hybrid model offers a convincing middle path: the ability to adapt artificial intelligence performance based on the task, from immediate customer support response to complex financial evaluation. The system maintains anthropics Previous prices with 3 USD for one million input tokens and USD 15 for one million output tokens, even with additional reasoning functions.
“Our clients are trying to achieve results for their clients,” explained Michael Gerstenhaber, head of the anthropic platform. “Using the same model and monitors the same model in different ways allows someone the same Thompson Reuters To conduct legal examinations, it allows our partners corresponding to Cursor Or Girub To be able to develop applications and achieve these goals. “
Anthropic hybrid approach represents each technical and strategic evolution of Gambit. While OpenAI maintains separate models for various possibilities, and Deepseek focuses on Cost performanceAnthropic runs unified systems that may support each routine tasks and complex reasoning. It is a philosophy that might change the way firms implement artificial intelligence and eliminate the need for juggling with many specialized models.
Discover the Claude code: AI recent assistant
Today’s anthropic also presented Claude Codecommand line tool that enables programmers to delegate complex engineering tasks on to AI. The system requires man approval before making changes to the Code, reflecting the growing concentration of the industry on the responsible development of AI.
“In fact, you still have to accept the changes that Claude introduces. You are a reviewer with your hands [the] Wheel, “Penn noted. “There is basically a type of control list that should be generally accepted so that the model can take some actions.”
Ads appear among intensive competition in the development of artificial intelligence. Stanford researchers He recently created Open source reasoning model for lower than USD 50, while Microsoft has just integrated Model O3-Mini OPENAI on azure. Deepseek’s success has also stimulated a recent approach to the development of artificial intelligence, and some firms study model distillation techniques that might reduce costs much more.
From Pokémon to Enterprise: testing the recent intelligence AI
Penn illustrated the dramatic progress in the possibilities of AI with an unexpected example: “We asked various versions of Claude to play Pokémon … This version reached the end City of VermilionHe captured a lot of Pokémon and even grinds to the level. He has the right Pokémon to fight rivals. “
“I think we’ll see how innovations and press the quality of reasoning, press on such things as dynamic reasoning,” Penn explained. “We have always considered it the basic part of the intelligence, not something separate.”
The real test of the Anthropiku approach will come from the adoption of the company. While Pokémon could appear trivial, it shows the type of adaptive intelligence that the company needs: AI, which may support each routine operations and complex strategic decisions without switching between specialized models. Earlier Claude versions couldn’t move outside the initial city of the game. The latest version builds strategies, manages resources and makes tactical decisions-possibilities reflecting the complexity of real business challenges.
In the case of corporate clients, this may occasionally mean a difference between maintaining many AI systems for different tasks and the implementation of one, more talented solution. The following months will reveal whether the Antropica plant for unified reasoning of artificial intelligence will change the enterprise market, or will turn out to be one other experiment in the rapid evolution of the industry.
