// AI news · 2026-11-03 · Google Cloud AI/ML
Google Launches PLE Benchmark for Physical Engineering Agents
Google and Mercor launch PLE, a benchmark evaluating frontier AI agents' ability to perform real-world engineering tasks.
Published 2026-11-03 16:00 (Beijing time)Source Google Cloud AI/MLAuto-summarised, fact-checked against source
01Summary
Tasks include CAD models, BIM files, and engineering documents in four disciplines. Agents must generate compliant, safe designs passing strict safety checks. Current models score below 50% on private tests, with only 41.3% of outputs meeting all feasibility constraints.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- PLE evaluates agents on real-world engineering tasks across four disciplines.
- Tasks include CAD, BIM, and document files in realistic project environments.
- All frontier models score below 50% on private test set; only 41.3% pass all safety checks.
- Model choice has far greater impact than agent harness on performance.
03Source
Introducing Physical-Engineering’s Last Exam for agents
Google Cloud AI/ML · published 2026-11-03 16:00 Beijing time
05Related news
- Same source · Nammu21 Automates Syndicated Loans on GCUL2026-11-04 · Google Cloud AI/ML
- Same source · Google Cloud Launches General Availability of Spend Caps2026-10-31 · Google Cloud AI/ML
- Same topic · Cheap classifiers route decisions across Gemini models2026-11-03 · Google AI Changelog
- Same topic · Google Antigravity with Cheminformatics Skills speeds drug discovery2026-11-04 · Google DeepMind Blog
- Same day · Cities Build and Scale AI Services with NVIDIA2026-11-03 · NVIDIA Newsroom
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All stories of 2026-11-03 · Archive by date · All AI news · Today