# Google releases engineering tasks and tools for AI agent evaluation

_Published Tuesday, November 3, 2026 at 9:19 PM EST · AI · Latest · Tier 2 — Notable_

![Google releases engineering tasks and tools for AI agent evaluation — Primary](https://www.unite.ai/wp-content/uploads/2026/11/google-releases-physical-engineerings-last-exam-for-ai-agents.jpg)

Google AI & Infrastructure released 29 public engineering tasks, grading tools and an evaluation container on November 3, opening a competition for agents that produce professional engineering deliverables, Unite.AI reported. The release expands access to Physical-Engineering's Last Exam, an 85-task benchmark developed with Mercor.

Entrants submit the software that drives their chosen model through project files and produces documents such as calculations and drawings. Organizers test submissions on 56 hidden tasks. Google reports that every model scored below 50% on that hidden set using its native agent software. Hard safety or feasibility failures prevent affected designs from earning objective credit, even when they meet other requirements.

## Sources

- [Unite.AI](https://www.unite.ai/google-opens-85-task-engineering-benchmark-and-agent-competition/)

---
Canonical: https://techandbusiness.org/newswire/vwbMG3zfkmTR3At4JvtMCZ
Published: 2026-11-04T02:19:10.310Z
Story chronology: 2026-11-03T17:25:03.000Z
Retrieved: 2026-11-04T05:26:35.103Z
Publisher: Tech & Business (techandbusiness.org)