Research map / Vision-language and multimodal / Multimodal LLMs
VisWordBench: Bridging the Gap in Cross-modal Reasoning for Multimodal Large Language Models
Tianshu Zhang, Junzhe Chen, Yean Cheng, Demin Zhu, Haoze Zheng, Lijie Wen
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 |
|---|---|---|---|
| add inference-time steering from hypothesis-diversified data | GLM-4.1v-thinking · VisWordBench-en | accuracy (%): 3.43 -> 5.23 | Improved |
| add inference-time steering from hypothesis-diversified data | Kimi-vl-thinking · VisWordBench-en | accuracy (%): 1.66 -> 2.77 | Improved |
| add inference-time steering from hypothesis-diversified data | GLM-4.5v · VisWordBench-en | accuracy (%): 14.95 -> 16.18 | Improved |
| steer with conventional distillation data instead of our diversified data (ablation) | GLM-4.1v-thinking · VisWordBench-en | accuracy (%): 3.43 -> 4.43 | No clear effect |
| fine-tune on our diversified synthetic data instead of answer-only data (ablation) | GLM-4.1v-thinking · VisWordBench-en | accuracy (%): 5.23 -> 5.72 | Improved |
| fine-tune on our diversified synthetic data instead of answer-only data (ablation) | Kimi-vl-thinking · VisWordBench-en | accuracy (%): 2.52 -> 3.26 | Improved |
| LoRA on our diversified synthetic data instead of answer-only data (ablation) | GLM-4.5v · VisWordBench-en | accuracy (%): 16.12 -> 18.46 | Improved |
| raise steering strength alpha too high (ablation) | GLM-4.1v-thinking · VisWordBench-en | accuracy (%) | Hurt |
| apply steering vector from VisWordBench zero-shot | GLM-4.1v-thinking · VisuLogic | accuracy (%): 24.7 -> 25.9 | Improved |
| apply steering vector from VisWordBench zero-shot | GLM-4.1v-thinking · VisuRiddles | accuracy (%): 24.0 -> 26.7 | Improved |
| apply steering vector from VisWordBench zero-shot | GLM-4.1v-thinking · CrosswordBench | WCR_all (%): 6.04 -> 7.58 | Improved |
| enable thinking mode | Qwen3.5-9B · VisWordBench-ch | accuracy (%): 1.60 -> 2.20 | Improved |
Closest papers by meaning
- Evaluating Cross-Modal Reasoning Ability and Problem Charactaristics with Multimodal Item Response Theory (ICLR 2026 · similarity 0.93)
- Compose and Fuse: Revisiting the Foundational Bottlenecks in Multimodal Reasoning (ICLR 2026 · similarity 0.93)
- XModBench: Benchmarking Cross-Modal Capabilities and Consistency in Omni-Language Models (ICLR 2026 · similarity 0.92)
- CrossCheck-Bench: Diagnosing Compositional Failures in Multimodal Conflict Resolution (AAAI 2026 · similarity 0.92)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · similarity 0.92)
- Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability (NeurIPS 2026 · similarity 0.92)
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.
