Back to Project
M

MAGAR-Bench

Project Nov 5, 2026

A public test-input export for multimodal agents making 3D mechanical assembly decisions, GUI actions and bounded minimal-rework choices.

Visit primary link 4 sources attached

What happened lately

MAGAR-Bench timeline

The November 5, 2026 export distributes 1,000 test inputs spanning six assembly families. It contains prompts and image references, with images packaged separately, while training examples, gold answers and scoring remain outside the public release. Minimal-rework references are bounded by the validated candidate set, not unrestricted global optimality. The maintainer’s image-rights attestation remains pending. The documentation names an associated manuscript but supplies no verified public paper identifier.