Google and Mercor open PLE benchmark for physical-engineering agents
Physical-Engineering’s Last Exam evaluates production-like outputs from AI agents across 85 long-horizon tasks in mechanical, electrical, civil and structural engineering.
AT A GLANCE
Key points
- PLE contains 85 long-horizon tasks across four engineering disciplines.
- Google released 29 tasks, grading tools and an open competition.
- The highest private-set score was 48.3%, leaving expert review essential.
What is changing?
Google AI & Infrastructure and Mercor announced Physical-Engineering’s Last Exam (PLE) on November 3. The benchmark contains 85 long-horizon tasks across mechanical, electrical, civil and structural engineering, each running in a project folder with an average of 61 files.
Agents must inspect PDFs, emails, spreadsheets, CAD and BIM files, perform calculations and produce deliverables for review by a licensed engineer. A design earns cost-efficiency points only after passing every hard safety and feasibility check. Google released 29 public tasks, grading tools and an open competition environment.
In Google’s private evaluation, the strongest configuration reached a mean score of 48.3%. The result indicates that current models are not ready to perform engineering work independently and that expert oversight remains essential.
Explore the original sources
Source date: 3 November 2026.
Published by AI Radar: 4 Nov 2026, 03:07 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…