Dataset

Physical-Engineering’s Last Exam (Dev Set)

Contains realistic engineering task workspaces across 4 disciplines, with project files and tools for agents to produce engineering deliverables.

Hugging Face ↗Oct 18, 2026Paper ↗

  • llm
  • nlp
  • multimodal
  • benchmark
  • engineering-agents
  • long-horizon-tasks
Pipeline diagram showing the adversarial creator-solver co-curation loop method
Pipeline diagram showing the adversarial creator-solver co-curation loop method

What it is

Benchmark of long-horizon engineering tasks in realistic project workspaces across four disciplines, aimed at evaluating agentic engineering work. Dev-set benchmark of realistic engineering workspaces across 4 disciplines, with drawings, specs, spreadsheets, solver scripts, and CAD/BIM or simulation tools; tasks require producing deliverables like code, calculations, models, schedules, and reports.

Key results

  • 4 engineering disciplines
  • Includes drawings, specs, spreadsheets, and solver scripts
  • Uses CAD/BIM or simulation tools in the workspace

Useful for benchmarking agents on long-horizon, multi-tool engineering task execution.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.