Matching Object or Relation? Tracing Abstract Reasoning Inside VLMs
cs.AI
Submitted: 2026-10-06
Updated: 2026-10-06
License: http://creativecommons.org/licenses/by/4.0/
The gist: Vision Language Models (VLMs) excel on visual benchmarks but fail systematically on tasks requiring abstract reasoning.
Abstract
Vision Language Models (VLMs) excel on visual benchmarks but fail systematically on tasks requiring abstract reasoning. Existing benchmarks document this failure but cannot say why it happens or which cognitive capability is missing. We close this gap by adopting the Relational Match-to-Sample (RMTS) paradigm from comparative and developmental psychology and pairing it with a mechanistic analysis of the model's internals. On a parametrically controlled stimulus set evaluated across frontier API models (GPT, Claude, Gemini) and three open-source families (Qwen3.5, Gemma-4, InternVL3), we identify four levers that shift VLMs toward the relational match---capability tier, model scale, the number of objects per scene, and the absence of per-object stimulus noise---together producing a developmental-like trajectory that mirrors the human relational shift. Opening up the model, a per-layer representational similarity analysis and a causal mediation analysis reveal that VLM abstract reasoning is implemented by two competing circuits: an early circuit that organises images by their surface object features, and a late circuit that organises them by their abstract relation. Extending the analysis to ARC-AGI-1, we find that ablating the relational heads identified on RMTS degrades performance more than ablating random heads, indicating that the relational circuit is recruited beyond our controlled stimuli. We hope this mechanism-level view serves as a step toward understanding how abstract reasoning is implemented in VLMs.
Sources
- Visual symbolic mechanisms: Emergent symbol processing in vision language models
- Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas
- On the Measure of Intelligence
- ARC-AGI-2: A New Challenge for Frontier AI Reasoning Systems
- How to use and interpret activation patching
- Lost in Embeddings: Information Loss in Vision-Language Models
- Visual Large Language Models for Generalized and Specialized Applications
- A Tale of Two Circuits: Grokking as Competition of Sparse and Dense Subnetworks
- Bridging Lottery Ticket and Grokking: Understanding Grokking from Inner Structure of Networks
- Comparing Humans, GPT-4, and GPT-4V On Abstraction and Reasoning Tasks
- Slot Abstractors: Toward Scalable Abstract Visual Reasoning
- Qwen3.5-Omni Technical Report
- The Relational Bottleneck as an Inductive Bias for Efficient Abstraction
- IKOD: Mitigating Visual Attention Degradation in Large Vision-Language Models
- KiVA: Kid-inspired Visual Analogies for Testing Large Multimodal Models
- Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
- Multimodal Chain-of-Thought Reasoning in Language Models
- InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems