Research map / Vision-language and multimodal / Multimodal LLMs
How Do Multimodal Large Language Models Handle Complex Multimodal Reasoning? Placing Them in An Extensible Escape Game
Ziyue Wang, Yurui Dong, Fuwen Luo, Minyuan Ruan, Zhili Cheng, Chi Chen, Peng Li, Yang Liu
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 |
|---|---|---|---|
| increase difficulty from Difficulty-1 to Difficulty-3 | GPT-4o · MM-Escape single room | escape rate (%): 100.00 -> 71.36 | Hurt |
| increase difficulty from Difficulty-1 to Difficulty-3 | Gemini-1.5-pro · MM-Escape single room | escape rate (%): 81.82 -> 46.82 | Hurt |
| increase difficulty from Difficulty-1 to Difficulty-3 | Doubao 1.5 Pro · MM-Escape single room | escape rate (%): 91.91 -> 9.52 | Hurt |
| increase difficulty from Difficulty-1 to Difficulty-3 | Claude 3.5 Sonnet · MM-Escape single room | escape rate (%): 72.73 -> 39.61 | Hurt |
| provide successful path of Room 1 and use multi-room Difficulty-1 & Difficulty-2 | GPT-4o · MM-Escape Difficulty-2 | escape rate (%): 72.73 -> 90.00 | Improved |
| provide successful path of Room 1 and use multi-room Difficulty-2 & Difficulty-2 | Gemini-1.5-pro · MM-Escape Difficulty-2 | escape rate (%): 54.55 -> 60.00 | Improved |
| two Difficulty-1 rooms instead of one | GPT-4o · MM-Escape Difficulty-1 | escape rate (%): 100.00 -> 75.00 | Hurt |
| two Difficulty-1 rooms instead of one | Gemini-1.5-pro · MM-Escape Difficulty-1 | escape rate (%): 81.82 -> 22.22 | Hurt |
| compare intent-outcome consistency against grab success rate | GPT-4o · MM-Escape Difficulty-3 | percent: 26.51 vs 31.36 | Mixed |
| move the password pattern away from the door | GPT-4o · Extended scenario (qualitative) | escape | Hurt |
Closest papers by meaning
- MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods (NeurIPS 2026 · similarity 0.90)
- MMR-Life: Piecing Together Real-life Scenes for Multimodal Multi-image Reasoning (ICLR 2026 · similarity 0.90)
- Downscaling Intelligence: Exploring Perception and Reasoning Bottlenecks in Small Multimodal Models (CVPR 2026 · similarity 0.90)
- Play to Generalize: Learning to Reason Through Game Play (ICLR 2026 · similarity 0.89)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · similarity 0.89)
- Hyperphantasia: A Benchmark for Evaluating the Mental Visualization Capabilities of Multimodal LLMs (NeurIPS 2025 · similarity 0.89)
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.
