Anthropic, a prominent competitor to OpenAI, has announced a breakthrough in training its AI model, Claude, to independently operate computers. This includes searching the web, opening applications, and inputting text using a mouse and keyboard. While current AI models like OpenAI’s GPT-4 and Google’s Gemini excel in answering questions and holding human-like conversations, Claude’s ability to interact with computer systems sets it apart.
Claude surpasses other AI agents on key benchmarks such as OSWorld, which evaluates computer system usage efficiency. Notably, Claude successfully completes tasks in OSWorld 14.9% of the time, significantly surpassing GPT-4’s 7.7%. Anthropic’s progress mirrors other developments in agentic AI, such as Google’s Project Jarvis, which aims to develop AI agents that can autonomously execute tasks across devices.
These innovations reflect the growing potential of AI agents to handle complex, real-world operations, transforming how AI supports decision-making and productivity.