Anthropic’s new hybrid AI model can work on tasks autonomously for hours at a time

Claude Opus 4, designed for paying Anthropic customers, excels in handling complex tasks, while Claude Sonnet 4, available to both paid and free tiers, offers a more efficient approach for everyday use. Both hybrid models can switch between swift and reasoned responses based on the task's nature, drawing from web searches or other tools to refine outputs. AI companies are racing to create reliable, autonomous agents capable of planning, reasoning, and executing tasks without human oversight, though challenges like tool usage and safety remain. Anthropic has successfully reduced reward hacking in its new models by 65% through enhanced training methods and evaluation processes.