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

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

The company claims its ability to tackle complex, multistep problems paves the way for much more proficient AI agents.

technologyreview.com

TL;DR: Anthropic’s latest duo—Claude Opus 4 and Claude Sonnet 4—take AI agents to the next level. Opus 4 (for paying customers) can juggle multi-hour, multi-step tasks—think playing Pokémon Red for 24 hours straight to build a guide or coding autonomously for seven hours—by beefing up its “memory files.” The goal? Shift from hand-holding assistants to true agents that make key decisions on their own.

Sonnet 4 (available free and paid) and Opus 4 are both “hybrid” models, able to dial between quick responses and deep reasoning, even tapping the web or other tools mid-calculation. While Anthropic has trimmed down reward-hacking hiccups by about 65%, the broader race to build fully autonomous, safe AI agents still faces challenges around erratic behavior and unintended shortcuts.