Can LLMs beat video models? AA tests GPT-6.2 Astra vs Seedance 2.5
ArtificialAnlys · x · 2026-11-20
Artificial Analysis asks a contrarian question: can language models make better videos than video models? LLMs now produce motion graphics, edit video as code, and drive Blender to build 3D scenes.
- Subject: OpenAI's GPT-6.2 Astra (current #1 on AA-3D), building three scenes as three.js code — the Sydney Opera House, an exploded mechanical watch, and a dragon attacking a medieval town
- Method: scripted camera fly-through rendered in-browser at 15s, 1080p, 24fps
- Baseline: ByteDance Seed's Dreamina Seedance 2.5, given the same camera path as a text prompt with reference images, same settings
Results are shown side by side (LLM code scene on AA-3D vs video model on AA-Video-T2V), with interactive explorers and public arenas for readers to compare the two approaches themselves.
Related event: Artificial Analysis Tests LLM-Generated Video Against Video Models(2 posts)→
More from Models
- Google AI Mode now enforces usage limits, users surprised by rate-limit popup — elvano124 · 2026-11-22
- Sol 6.2 now refuses reverse engineering that Sol 6.1 happily did for a month — cardak98 · 2026-11-22
- Anthropic's record ~$81B IPO, GPT-6.2 pricing, Qwen4-Max and Copilot Autopilot: week in review — ksraj1001 · 2026-11-22
- Reddit rant: Frontier AI labs are training models to imitate lazy engineers — RossPeili · 2026-11-22
- MBZUAI open-sources PAN 2, a 29B-parameter general-purpose world model — LinusEkenstam · 2026-11-22
- Inkling Mini turns out surprisingly fluent in Russian, sparking data recipe questions — teortaxesTex · 2026-11-22