Part 7/12:
One of the most fascinating metrics shared was GPT-5's ability to perform autonomous tasks over longer periods. A graph plotting success rates at varying timeframes revealed exponential growth in the model’s capacity to manage tasks lasting minutes, now reaching over 26 minutes on average at a 50% success rate. The trajectory suggests that within a year, models may reliably handle autonomous tasks spanning hours or even longer, edging closer to human-level endurance in execution.