GPT-5.6-Sol Autonomously Completes Voxel-Based Manhattan Project Over One Week
The new GPT-5.6-Sol model successfully generated a voxel-based Manhattan, operating autonomously for nearly a week to achieve the task with high precision. This performance aligns with METR's time-horizon graph, suggesting significant progress in AI model capabilities and autonomy.
So What
A frontier model demonstrating week-long autonomous execution on a complex task signals a new level of agentic capability and sustained compute demand.