A new report from the nonprofit METR shows that recent AI models can complete tasks that would take a human a full work week. The findings update a widely used graph of AI progress that tracks capabilities using a time horizon measurement plot. The study evaluated a range of models to see how quickly they could finish complex workflows.

The report highlights that AI capabilities continue to double every 3.6 months. It notes that models are currently better at coding than at tasks requiring judgment. OpenAI’s GPT-6 Astra is included among the models evaluated by METR. The time horizon plot visualizes how long it takes for an AI to complete a task compared to a human.

This rapid acceleration suggests a fundamental shift in how software teams can operate. The gap between coding ability and judgment tasks remains a key area for future development. Developers and product managers should prepare for workflows that rely heavily on automated coding agents. The pace of progress means tools will become significantly more autonomous in the coming months.