Physics Benchmark for Video World Models: Best of 8 SOTA Scores Just 57.76/100
nikola_mr64990 · x · 2026-10-09
EinsiaAI introduced World Models' Last Exam in Physics, a SWE-bench-style benchmark for physical consistency in video world models.
- Evaluates 8 SOTA video world models on 40 controlled tasks across 9 physics categories, all scored via direct programmatic measurement against real physical laws.
- Top score: 57.76/100. Notably, free-fall videos passed the consistency screen but averaged just 26.61/100 composite — a large gap to close.
- Authors argue whoever solves this gap could shape the next generation of trustworthy world models.
Paper, GitHub, and project page are publicly available.
More from Research
- Claim-Locked Reporting: EMNLP paper fixes LLM hallucinated claims over correct numbers — jiqizhixin · 2026-10-09
- OpenAI releases 372 math proof claims, including 23 Erdős problems spanning 1268 pages — burny_tech · 2026-10-09
- NVIDIA NeMo-DCR cuts 1T-model RL weight sync from 87.5 min to 150 sec — IanAndrewsDC · 2026-10-09
- Dex-One2Many: one human video trains a dexterous hand that transfers zero-shot to real robots — furongh · 2026-10-09
- GenIA: Rendering-Guided Test-Time Alignment Turns SAM3D into SOTA Image-to-3D — JonathonLuiten · 2026-10-09
- TIER IV open-sources METEOR: one 54M-parameter net runs 12 driving tasks at 15 FPS on Jetson — rsasaki0109 · 2026-10-09