Paper Detail
Kun Xiang, Terry Jingchen Zhang, Zirong Liu, Bokai Zhou, Yueling Tang, Junjie Yu, Jiacong Lu, Shangrui Huang, Heng Li, Likui Zhang, Kunkun Liu, Changzheng Zhang, Yangle Fang, Boqiang Guo, Hui-Ling Zhen, Dandan Tu, Yinya Huang, Xiaodan Liang
We introduce SeePhys Pro, a fine-grained modality transfer benchmark that studies whether models preserve the same reasoning capability when critical information is progressively transferred from text to image. Unlike standard vision-essential benchmarks that evaluate a single input form, SeePhys Pro features four semantically aligned variants for each problem with progressively increasing visual elements. Our evaluation shows that current frontier models are far from representation-invariant reasoners: performance degrades on average as information moves from language to diagrams, with visual variable grounding as the most critical bottleneck. Motivated by this inference-time fragility, we further develop large training corpora for multimodal RLVR and use blind training as a diagnostic control, finding that RL with all training images masked can still improve performance on unmasked validation sets. To analyze this effect, text-deletion, image-mask-rate, and format-saturation controls suggest that such gains can arise from residual textual and distributional cues rather than valid visual evidence. Our results highlight the need to evaluate multimodal reasoning not only by final-answer accuracy, but also by robustness under modality transfer and by diagnostics that test whether improvements rely on task-critical visual evidence.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{xiang2026seephys,
title = {SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning},
author = {Kun Xiang and Terry Jingchen Zhang and Zirong Liu and Bokai Zhou and Yueling Tang and Junjie Yu and Jiacong Lu and Shangrui Huang and Heng Li and Likui Zhang and Kunkun Liu and Changzheng Zhang and Yangle Fang and Boqiang Guo and Hui-Ling Zhen and Dandan Tu and Yinya Huang and Xiaodan Liang},
year = {2026},
abstract = {We introduce SeePhys Pro, a fine-grained modality transfer benchmark that studies whether models preserve the same reasoning capability when critical information is progressively transferred from text to image. Unlike standard vision-essential benchmarks that evaluate a single input form, SeePhys Pro features four semantically aligned variants for each problem with progressively increasing visual elements. Our evaluation shows that current frontier models are far from representation-invariant re},
url = {https://huggingface.co/papers/2605.09266},
keywords = {multimodal reasoning, representation-invariant reasoners, modality transfer, vision-essential benchmarks, multimodal RLVR, blind training, visual variable grounding, image-mask rate, text-deletion, format-saturation, code available, huggingface daily},
eprint = {2605.09266},
archiveprefix = {arXiv},
}
{}