This article explores measurement methods for AI systems’ ability to complete long-term tasks and showcases performance data from the Opus 4.5 system. Research findings indicate that Opus 4.5 has a 50% probability of completing tasks lasting up to 4 hours and 49 minutes, representing a significant advancement in current AI technology for long-term task processing. This capability assessment is crucial for AI applications requiring extended operation, such as autonomous driving, complex problem-solving, and continuous monitoring systems. The article provides detailed descriptions of testing methods and evaluation criteria, offering new approaches for the objective assessment of AI capabilities.
