Anthropic’s Claude 3.7 Sonnet takes aim at OpenAI and DeepSeek in AI’s next big battle

Anthropic I just drew a warning OPENAI,, In depth and the whole AI industry with the launch of Claude 3.7 SonnetA model that gives users unprecedented control over the time that AI devotes an “reflection” before generating an answer. The exit, alongside the beginnings of Claude CodeAn AI coding agent of command line, indicates the aggressive thrust of anthropic on the company’s AI market – a push that could reshape the way companies create software and automate work.

The stakes could not be higher. Last month, Deepseek amazed the world of technology with an AI model that corresponded to the capacities of American systems to a cost fractionNVIDIA sending Stock down 17% And the mastery of alarms on the American leadership of AI. Now, Anthropic is betting that a specific control over the reasoning of the AI ​​- not just raw speed or cost savings – will give it an advantage.

“We just think that reasoning is a fundamental part and a central component of an AI, rather than a separate thing that you have to pay separately to access,” said Dianne Penn, who directs product management at research at research at research at research at research at research Anthropic, in an interview with Venturebeat. “Like humans, AI should manage fast responses and complex thought. For a simple question as “what time is it?”, He should answer instantly. But for complex tasks – such as planning a two -week trip to Italy while meeting gluten -free food needs – it needs a more important processing time. »»

“We do not see reasoning, planning and self-correction as separate capacities,” she added. “It is therefore essentially our way of expressing this philosophical difference … Ideally, the model itself should recognize when a problem requires more intensive reflection and adjustment, rather than asking users to explicitly select different modes of reasoning.”

The reference data support the ambitious vision of anthropic. In extended mode of reflection, Claude 3.7 Sonnet reaches Precision of 78.2% On reasoning tasks at graduates, questioning the latest OpenAi and surpassing Deepseek-R1 models.

But the most revealing measures come from real world applications. Model scores 81.2% on the use of tools focused on retail and shows marked improvements in according to instructions (93.2%) – Domains where competitors had trouble or have not published results.

While Deepseek and Openai lead Traditional mathematical benchmarksClaude 3.7’s unified approach demonstrates that a single model can effectively switch between rapid responses and deep analysis, potentially eliminating the need for companies to maintain separate AI systems for different types of tasks.

How the hybrid anthropic could reshape the computer company

The moment of release is crucial. The emergence of Deepseek sent last month shock waves Through the Silicon Valley, demonstrating that the sophisticated AI reasoning could be made with much less computing power than before. This challenged fundamental hypotheses concerning AI development costs and infrastructure requirements. When Deepseek published its results, Nvidia’s actions dropped by 17% In a single day, investors suddenly wondering if expensive fleas were really essential for advanced AI.

For companies, the challenges could not be higher. Companies are Spend millions Integration of AI into their operations, Paris on which approach will dominate. The hybrid model of Anthropic offers a convincing average path: the possibility of adjusting the performance of the AI ​​based on the task to be accomplished, from instantaneous customer service responses to a complex financial analysis. The system maintains anthropes previous price $ 3 per million entry tokens and $ 15 per million output tokens, even with additional reasoning features.

“Our customers are trying to obtain results for their customers,” said Michael Gerstenhaber, head of the Anthropic platform. “Using the same model and inviting the same model in different ways allows someone as Thompson Reuters doing legal research, allows our coding partners as Cursor Or Github To be able to develop applications and achieve these objectives. »»

The hybrid approach of anthropic represents both a technical evolution and a strategic gambit. While Optaai retains separate models for different capacities and Deepseek focuses on profitabilityAnthropic pursues unified systems which can manage both routine tasks and complex reasoning. It is a philosophy that could reshape the way companies deploy AI and eliminate the need to juggle several specialized models.

Meet the Code Claude: the new AI developer

Anthropic today also revealed Claude CodeA command line that allows developers to directly delegate complex engineering tasks to AI. The system requires human approval before committing code changes, reflecting the concentration of growing industry on responsible development of AI.

“You must always accept the changes that Claude brings. You are a critic with [the] wheel, “noted Penn. “There is essentially a kind of control list that you must essentially accept so that the model takes certain actions.”

The announcements are involved in the midst of intense competition in the development of AI. Stanford researchers recently created a Open Source reasoning model For less than $ 50, while Microsoft has just integrated it O3-Mini of Openai model In Azure. Deepseek’s success has also stimulated new AI development approaches, some companies exploring the model’s distillation techniques that could further reduce costs.

From Pokémon to the company: Testing the new intelligence of AI

Penn illustrated the dramatic progress of the AI ​​capabilities with an unexpected example: “We asked different versions of Claude to play Pokémon … This version made Vermion CityCaptured several Pokémon and even grinks at level. He has the right Pokémon to fight against rivals. »»

“I think you will see us continue to innovate and push the quality of reasoning, push towards things like dynamic reasoning,” said Penn. “We have always considered it a fundamental part of intelligence, rather than something separate.”

The real test of Anthropic’s approach will come from the adoption of companies. Although playing Pokémon may seem trivial, it demonstrates the type of adaptive intelligence that companies need: IA that can manage both routine operations and complex strategic decisions without switching between specialized models. The previous versions of Claude could not navigate beyond the departure city of a game. The latest version develops strategies, manages resources and makes tactical decisions – capacities that reflect the complexity of the business challenges of the real world .

For business customers, this could make the difference between the maintenance of several AI systems for different tasks and the deployment of a unique and more competent solution. The coming months reveal whether the bet of anthropic on the reasoning of unified AI will reshape the business market or just become another experience in the rapid evolution of the industry.