one dataset · paper, tables, graph, wiki
simgrammar the grammar of simulation

locator

https://arxiv.org/abs/2508.13009

notes

25 fps on one H100 at 352x640; 1.3B base plus action modules (~1.8B); ~1,200 h from Unreal Engine, Minecraft, GTA driving, Temple Run, Sekai; Self-Forcing with DMD, 3 steps, rolling KV cache; weights and code released; no explicit memory.

cited by

1 records

Open real-time interactive video world models

provenance

runs in log/runs
2026-10-24
created2026-10-24-techniques-world-modelstechnique