Thought Leadership
Anthropic has unveiled two new AI models that can perform complex tasks autonomously for hours on end. The company sees this as a paradigm shift from simple chatbots to autonomous work agents.
With Claude Opus 4 and Claude Sonnet 4, the AI company Anthropic is expanding its model family with two systems that it claims will set new standards in autonomous task processing. According to the manufacturer, the models can analyze large amounts of data, carry out time-consuming work steps independently and remain focused for several hours.
Change of strategy: away from chatbots and towards work agents
Anthropic Chief Scientist Jared Kaplan explains the company’s strategic change of course: since the end of 2023, Anthropic has no longer been investing primarily in chatbot functionalities, but is instead focusing on the development of AI systems for complex workflows. This includes extensive research, programming tasks and even the creation of entire software projects.
Kaplan sees stability as the biggest challenge: “With more complex tasks, the risk of the model going off course increases.” The development team is working hard to solve this problem so that users can safely delegate extensive work orders.
Technical innovations and working methods
According to the manufacturer, Claude Opus 4 can work continuously on a task for up to seven hours – almost the duration of a full working day. Anthropic advertises the system as the “world’s best programming model”.
The new models have web search functions and can switch between analytical processes and the use of external tools. When accessing local file systems, they store relevant information temporarily in order to document work progress and continuously build up knowledge.
Mike Krieger, Head of Product at Anthropic, describes his personal experience with the new models: While he mainly used earlier versions of Claude as a “thinking partner” and wrote most of the texts himself, Claude Opus 4 now takes over the majority of his writing work. The quality is said to be so high that the texts can no longer be distinguished from his own. If this is true, Claude will remain one of the biggest competitors for OpenAI and ChatGPT.
