Claude Opus 4.7 Beats Human Teams 10-37x in Robodog Tasks, Still Misses Final Fetch
Updated
Updated · Anthropic · Jun 18
Claude Opus 4.7 Beats Human Teams 10-37x in Robodog Tasks, Still Misses Final Fetch
2 articles · Updated · Anthropic · Jun 18
Summary
Claude Opus 4.7 completed every robodog task that at least one human team had finished in Project Fetch, doing so at least 10 times faster and about 20 times faster than the best human team overall.
Across the four tasks both human teams completed, the model averaged more than 37 times the speed of the Claude-less team and more than 18 times that of the Claude-assisted team, while generating nearly 10 times less code than Team Claude.
Anthropic said the model quickly chose workable sensor interfaces and often wrote effective code on the first try, with a researcher limited to connecting the laptop, entering the prompt and approving commands.
The gains did not extend to the hardest step: Opus 4.7 still struggled with the closed-loop control needed to precisely move a beach ball and could not autonomously complete the final fetching task.
Anthropic framed the result as evidence that general model scaling is pushing AI toward limited physical agency, while cautioning that low-level robotic control and broader hardware adaptability remain unresolved.
Why won't Anthropic fix a flaw that lets hackers control its new AI-powered robots?
As AI becomes 37x faster than humans, are we prepared for its catastrophic dual-use risks?
AI now wins real-world deals for humans. What happens when it controls physical labor next?
Anthropic’s Claude Opus 4.7 Sets New Robotics Benchmark: Project Fetch in 1h15m
Overview
Anthropic's Claude Opus 4.7 set a new standard in robotics by leading a robotic dog to complete the Project Fetch challenge in June 2026. This event tested robotic autonomy in complex physical tasks, and Claude Opus 4.7 finished in just 1 hour and 15 minutes—nearly three times faster than the average human team, which took 3 hours and 40 minutes. The AI's remarkable speed and efficiency highlight a major leap in AI's ability to handle real-world physical challenges, clearly outpacing human performance and showcasing the future potential of autonomous systems.