BVB benchmark tests agentic video understanding by rebuilding real videos in Blender
RexDouglass · x · 2026-10-07
BVB (Blender-VideoBench) is a new open-source benchmark built on a simple premise: an agent that truly understands a video should be able to reconstruct it programmatically. Models must rebuild real-world videos as animated Blender scenes through the same lightweight harness (Mini-BVB), in an identical sandbox under the same cost limit, with no external asset libraries allowed. The leaderboard already compares GPT-6 variants and Claude Sonnet 5.5 / Opus 5.5, and the repo is MIT licensed.
More from Multimodal
- Open-source Claude skill huashu-art-motion: AI-made science video hits 100K views on Douyin in 5 hours — AlchainHust · 2026-10-07
- AI music video puts different AI models head-to-head — DeathCharge5 · 2026-10-07
- Open-source art animation skill: 35 styles, 9 explainer grammars, runs in Claude Code — AlchainHust · 2026-10-07
- Creator Says His AI Video Was Fully 1-Shot With Revid — Zero Edits, But Not Cheap — tibo_maker · 2026-10-07
- Midjourney v8.2 Imaginary Tokens: 4 Made-Up Words, 4 Surreal Worlds — LudovicCreator · 2026-10-07
- AI-animated series 'Gods Don't Give Gifts' reveals 20+ person team, 9 months of work — TomLikesRobots · 2026-10-07