Claude Opus 5.5 finished Baldur's Gate 3 on its own in 39 hours (bogdantruta.com)

Claude Opus 5.5, an AI-powered agent, recently completed the popular role-playing game Baldur's Gate 3 in just 39 hours, demonstrating its capabilities to autonomously navigate complex game mechanics and character development. Utilized in a unique setup involving a Linux build and modding tools, the AI controlled a half-orc paladin named Claude through 96 sessions, notably completing intricate tasks such as managing turn-based combat and character interactions without any human intervention. The final victory against the game's ultimate antagonist, the Netherbrain, showcased the AI's strategic planning and adaptability as it learned to overcome diverse challenges and character decisions. This achievement is significant for the AI/ML community as it illustrates the potential of AI in engaging with and mastering complex narrative-driven games, setting a precedent for future developments in gaming AI. Technically, Claude Opus 5.5 strategically made 20,809 tool calls, with 12% encountering errors, while employing innovative methods to solve in-game puzzles and navigate obstacles, such as using Light spells to counteract magical blindness. Moreover, the AI's self-policing behavior, where it honored game mechanics like a paladin's oath, exemplifies an evolving intelligence capable of nuanced decision-making in dynamic environments, further blurring the lines between human and machine gameplay.