VibePrevisBench: a benchmark turning scripts, photos or videos into executable Blender 4D scenes
_akhaliq · x · 2026-11-16
VibePrevisBench is a new benchmark that evaluates coding agents by having them turn scripts, photos, or videos into executable 4D scenes in Blender. Scoring combines a rule-based scene check with a VLM-based check of the proxy video. The pipeline shows an agent reconstructing the 4D scene, rendering a proxy video, then handing appearance off to a real-world video generation model; weak vs strong agent comparisons show clear score gaps.
More from coding & agent
- Nebius analyzes 143K coding-agent trajectories: equal solve rates, very different behaviors — nebius · 2026-11-16
- Perplexity agents can cite every number in financial research to the exact filing line item — AravSrinivas · 2026-11-16
- Weaviate paper: tiered retrieval effort lets you trade accuracy, latency and cost per query — CShorten30 · 2026-11-16
- Reranking study: Zerank-2 + LLM second pass hits 0.47 nDCG@10 at ~$0.037/query — CShorten30 · 2026-11-16
- Dozens of MCP tool schemas eat your agent's context — defer with alwaysLoad — Independent-Sir3234 · 2026-11-16
- PyBot: a chatbot with no weights, 600k lines of hand-written Python matching SmolLM 360M — anton_lozhkov · 2026-11-16