Independent AI research
Learn what AI can build by asking it to build.
We experiment with models in real creative and engineering tasks. Our current focus is 3D game development: a demanding way to study code, interaction, visual structure, and iteration together.
Read our articlesResearch in development
3D game development benchmarks
We are designing the tasks and evidence standard. No public model rankings are claimed for this work yet.
How we intend to evaluate
A playable result is the evidence.
The brief
Give models a comparable 3D game task, constraints, and a definition of a working result.
The build
Record code, assets, tool use, revisions, time, and points where human guidance was needed.
The playtest
Check whether the game runs, responds correctly, communicates its rules, and can be improved.
Model measurement continues
A broader lab, built on evaluation.
Earlier work on model quality, cost, speed, and workflows remains available. It informs the way we will study games and apps, while our main focus shifts toward building them.