This is the most misunderstood graph in AI

MITs nonprofit research group METR (Model Evaluation & Threat Research) has updated its influential graph tracking AI capabilities, revealing that Anthropics latest large language model, Claude Opus 4.5, significantly outperforms previous trends by potentially completing tasks that would take humans around five hours, far exceeding prior exponential growth predictions. However, METR cautions that these performance estimates have wide uncertainty ranges, with Opus 4.5s true capabilities possibly corresponding to tasks requiring anywhere from two to 20 human hours, highlighting both the rapid advancement and the complexity of accurately assessing AI progress.