4DCodeBench tests coding agents' motion reconstruction

Coding agents struggle to rebuild motion from video, 4DCodeBench reveals.

By ·

Oct 11, 2026

3 cards · 40s · tap to begin

From · · · 40s

4DCodeBench tests coding agents' motion reconstruction

Coding agents struggle to rebuild motion from video, 4DCodeBench reveals.

In brief

Coding agents struggle to rebuild motion from video, 4DCodeBench reveals. Stanford PhD student Ziga Kovacic's 4DCodeBench benchmark shows coding agents can't reconstruct motion from video. Originally reported by RuntimeWire.

4DCodeBench benchmark

Stanford PhD student Ziga Kovacic introduced 4DCodeBench to test coding agents' ability to turn video into executable graphics code.

4DCodeBench finds motion gap

Coding agents produce plausible geometry without reconstructing motion that governs a scene, according to the benchmark.

4DCodeBench paper published

4DCodeBench paper published

The paper appeared on arXiv on October 2nd; Kovacic described the project on X on October 5th.

What matters most

Stanford PhD student Ziga Kovacic's 4DCodeBench benchmark shows coding agents can't reconstruct motion from video.

Read the original on RuntimeWire

React

Sign in to react and comment.

Comments (0)

Life is short. Keep it sweet. Respect others' opinions and be kind!

    Recommended next

    More decks on 4dcodebench and related topics.