# The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images

arXiv: https://arxiv.org/abs/2608.06270 · HTML: https://mlsanity.com/paper/2608.06270/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 14 of its 31 references that are in the map.
- First posted on arXiv 6 Aug 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, 14 of 14)

- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [Measuring Faithfulness in Chain-of-Thought Reasoning](https://mlsanity.com/paper/2307.13702/) (2023)
- [DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning](https://mlsanity.com/paper/2505.14362/) (2025)
- [Pixel Reasoner: Incentivizing Pixel-Space Reasoning with Curiosity-Driven Reinforcement Learning](https://mlsanity.com/paper/2505.15966/) (2025)
- [Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers](https://mlsanity.com/paper/2506.23918/) (2025)
- [Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search](https://mlsanity.com/paper/2509.07969/) (2025)
- [Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception](https://mlsanity.com/paper/2602.11858/) (2026)
- [Chain-of-Thought Unfaithfulness as Disguised Accuracy](https://mlsanity.com/paper/2402.14897/) (2024)
- [CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization](https://mlsanity.com/paper/2511.19661/) (2025)
- [DeFacto: Counterfactual Thinking with Images for Enforcing Evidence-Grounded and Faithful Reasoning](https://mlsanity.com/paper/2509.20912/) (2025)
- [From Illusion to Intention: Visual Rationale Learning for Vision-Language Reasoning](https://mlsanity.com/paper/2511.23031/) (2025)
- [What Does Vision Tool-Use Reinforcement Learning Really Learn? Disentangling Tool-Induced and Intrinsic Effects for Crop-and-Zoom](https://mlsanity.com/paper/2602.01334/) (2026)
- [TACO: Tool-Augmented Credit Optimization for Agentic Tool Use](https://mlsanity.com/paper/2606.30251/) (2026)
- [CauSight: Learning to Supersense for Visual Causal Discovery](https://mlsanity.com/paper/2512.01827/) (2025)

## Built on by (most cited first, 2 of 2)

- [Aphanta: Diagnosing Task-Aligned Image-Edited Intermediates for Multimodal Reasoning](https://mlsanity.com/paper/2608.26993/) (2026)
- [When Should a VLM Look? Paying Only for Visual Calls That Were Needed and Used](https://mlsanity.com/paper/2609.22910/) (2026)

## Abstract (arXiv)

The "thinking-with-images" paradigm equips multimodal LLMs with active visual operations such as crop-and-zoom. However, models using these operations often achieve only marginal or negative gains over direct inference at substantially higher token cost. They may also repeatedly crop irrelevant regions and fail on questions that direct inference answers correctly. We ask whether the returned visual evidence causally affects the answer. To answer this question, we formulate visual tool-use as a causal graph that separates observation-mediated paths from action-induced shortcuts. We then audit it through interventions at the three levels: policy (comparing tool-use with direct inference), trajectory (corrupting all observations during rollout), and step (counterfactually replacing one individual observation under a fixed prefix). Our step-level estimand, Visual Evidence Gain, isolates the contribution of each returned observation. Across six representative models and five fine-grained perception benchmarks, we uncover policy miscalibration with two failure modes. In Calling Without Looking, returned observations have no causal effect on the answer. In Looking Without Planning, observations are informative but the call schedule is incoherent. A trajectory-level diagnostic decomposes the policy-level accuracy gain into per-group contributions and shows that the gain is concentrated in a Calibrated minority. We term this discrepancy the illusion of visual tool-use: despite aggregate accuracy gains, visual tool-use is not causally effective across a broad range of rollouts. The code is available at https://github.com/OpenCausaLab/CauAudit.