# From Perception to Integration: Revisiting the Internal Dynamics of Reasoning in Vision-Language Models

arXiv: https://arxiv.org/abs/2609.34809 · HTML: https://mlsanity.com/paper/2609.34809/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 22 of its 38 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 22 of 22)

- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [Are We on the Right Way for Evaluating Large Vision-Language Models?](https://mlsanity.com/paper/2403.20330/) (2024)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [Multimodal Chain-of-Thought Reasoning in Language Models](https://mlsanity.com/paper/2302.00923/) (2023)
- [Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning](https://mlsanity.com/paper/2403.16999/) (2024)
- [Dynamic Early Exit in Reasoning Models](https://mlsanity.com/paper/2504.15895/) (2025)
- [How to use and interpret activation patching](https://mlsanity.com/paper/2404.15255/) (2024)
- [Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification](https://mlsanity.com/paper/2504.05419/) (2025)
- [MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency](https://mlsanity.com/paper/2502.09621/) (2025)
- [Towards Interpreting Visual Information Processing in Vision-Language Models](https://mlsanity.com/paper/2410.07149/) (2024)
- [Cross-modal Information Flow in Multimodal Large Language Models](https://mlsanity.com/paper/2411.18620/) (2024)
- [Understanding Information Storage and Transfer in Multi-modal Large Language Models](https://mlsanity.com/paper/2406.04236/) (2024)
- [VoCoT: Unleashing Visually Grounded Multi-Step Reasoning in Large Multi-Modal Models](https://mlsanity.com/paper/2405.16919/) (2024)
- [Automatic Generation of Contrast Sets from Scene Graphs: Probing the Compositional Consistency of GQA](https://mlsanity.com/paper/2103.09591/) (2021)
- [Knowing Before Saying: LLM Representations Encode Information About Chain-of-Thought Success Before Completion](https://mlsanity.com/paper/2505.24362/) (2025)
- [Probing Conceptual Understanding of Large Visual-Language Models](https://mlsanity.com/paper/2304.03659/) (2023)
- [Human Cognitive Benchmarks Reveal Foundational Visual Gaps in MLLMs](https://mlsanity.com/paper/2502.16435/) (2025)
- [How Multimodal LLMs Solve Image Tasks: A Lens on Visual Grounding, Task Reasoning, and Answer Decoding](https://mlsanity.com/paper/2508.20279/) (2025)
- [When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoning](https://mlsanity.com/paper/2604.06787/) (2026)
- [Pathways of Visual Information Flow in Vision-Language Models](https://mlsanity.com/paper/2607.03358/) (2026)
- [Look Light, Think Heavy: What Multimodal Chain-of-Thought Reasoning Can and Cannot Do](https://mlsanity.com/paper/2606.22565/) (2026)

## Abstract (arXiv)

Vision-language models (VLMs) can answer simple visual questions, but often struggle when one question requires several visual judgments. We study this gap with controlled tasks for feature binding, numerosity, spatial relations, and amodal completion, together with a Composite task that combines them. Matched counterfactual image pairs isolate changes in the visual evidence needed to answer. Across four models, direct answers, hidden-state readouts, and state interventions show that the individual judgments can be made without explicit reasoning and that intervening on the corresponding states can affect the answer. During reasoning, the Composite answer becomes decodable from hidden states and usable from shortened traces, often before the model stops on its own. We train a small detector to predict this readiness and stop reasoning at that point. On MMStar and RealWorldQA, this reduces mean reasoning tokens by 79.1% and 74.5%, while average accuracy rises by 3.13 and 3.30 percentage points, respectively. These findings connect the internal development of answer readiness to a practical rule for allocating reasoning computation.