Anthropic's Claude Opus 4.7 completed complex robot tasks in just 9 minutes that previously required humans 181 minutes to finish, marking a dramatic acceleration in AI-powered robotics capabilities.

The breakthrough emerged from Anthropic's Project Fetch research, which tracked how increasingly powerful foundation models improve real-world robot performance without specific robotics training.

In August 2025, Claude Opus 4.1 proved completely unable to control a quadruped robot autonomously. Humans working alongside the model were only twice as effective as those without AI assistance, taking over three hours to complete the task set.

By May 2026, Opus 4.7 operating independently finished all tasks except one in 9 minutes and 35 seconds. The model failed only at repositioning a ball it had hit — a task that also challenged human operators.

"This progress is not the result of a concerted effort to improve the robotics capabilities of our models," Anthropic wrote. "These improvements have emerged from much more general scaling."

The bitter lesson applies to robotics

The results align with findings from robot startup Sunday, which reported that scaling pre-trained models paired with minimal high-quality data produces the most reliable robots.

Sunday's ACT-2 model achieved a 99.1% success rate across 778 clothing-folding tasks, handling nine different garment types. The company attributed success to "scale pretraining, then hill-climb with minimal in-house data."

"As the pretrained model becomes stronger, gains learned from a small amount of in-house data become increasingly transferable," Sunday explained.

Simple garments like shorts and t-shirts proved easiest for the robots, while complex items like blouses remained challenging despite maintaining above 90% success rates.

The research suggests that robotics breakthroughs may emerge as natural dividends of foundation model scaling rather than requiring specialized robotic training approaches. Most consumer robots remain limited by brittleness and poor generalization — problems that more capable base models could solve.

Sunday plans to deploy its folding robots to families through a beta program this fall, while Anthropic continues developing its general-purpose models.