Research map / Vision-language and multimodal / Multimodal LLMs
How Far Are Vision-Language Models from Constructing the Real World? A Benchmark for Physical Generative Reasoning
Luyu Yang, Yutong Dai, An Yan, Viraj Prabhu, Ran Xu, Zeyuan Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| switch Gemini from Planner-Atomic to Planner-Managed protocol | Gemini 3 Flash · DreamHouse Frame | structural pass rate: 45.4% -> 78.5% | Improved |
| switch GPT-5 from Planner-Atomic to Planner-Managed | GPT-5 · DreamHouse Frame | structural pass rate: 79.2% -> 33.3% | Hurt |
| switch Claude from Planner-Atomic to Planner-Managed | Claude Opus 4.5 · DreamHouse Frame | structural pass rate: 71.6% -> 71.3% | No clear effect |
| occlude framing with facade cladding | GPT-5 · DreamHouse Planner-Atomic | structural pass rate: 79.2% -> 63.7% | Hurt |
| occlude framing with facade cladding | Claude Opus 4.5 · DreamHouse Planner-Atomic | structural pass rate: 71.6% -> 77.9% | Improved |
| add up to 10 visual-feedback revisions (ablation) | Claude Opus 4.5 · DreamHouse refinement | mean visual score: 0.408 -> 0.441 | Improved |
| add up to 10 visual-feedback revisions (ablation) | Gemini 3 Flash · DreamHouse refinement | mean visual score: 0.433 -> 0.443 | Improved |
| add per-step visual feedback to Planner-Managed (ablation) | Claude Opus 4.5 · DreamHouse Frame | joint pass rate: 0.031 -> 0.094 | Improved |
| add per-step visual feedback to Planner-Managed (ablation) | GPT-5 · DreamHouse Frame | joint pass rate: 0.008 -> 0.009 | No clear effect |
| add per-step visual feedback to Planner-Managed (ablation) | Gemini 3 Flash · DreamHouse Frame | structural pass rate: 0.785 -> 0.740 | Hurt |
| use open-source models instead of frontier | Qwen3-VL / Kimi K2.5 · DreamHouse Planner-Managed | structural pass rate: 0% vs 78.5% (Gemini) | Hurt |
Closest papers by meaning
- GGBench: A Geometric Generative Reasoning Benchmark for Unified Multimodal Models (CVPR 2026 · similarity 0.90)
- Vision Language Models Cannot Reason About Physical Transformation (ICML 2026 · similarity 0.88)
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.88)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.88)
- PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (ICLR 2025 · similarity 0.88)
- QUANTIPHY: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models (CVPR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
