V-Thinker: Interactive Thinking with Images

TL;DR

V-Thinker tackles the challenge of teaching vision-centric, interactive thinking to large multimodal models. It introduces a Data Evolution Flywheel to auto-generate diverse, high-quality interactive data and a Visual Progressive Training Curriculum to align perception with interactive reasoning, complemented by VTBench for expert-verified evaluation. Empirical results show consistent gains over strong baselines in both general and interactive reasoning tasks, and the flywheel demonstrates scalable knowledge expansion across domains. This work provides a data-driven pathway toward robust, image‑interactive reasoning in multimodal systems with broad practical implications.

Abstract

Empowering Large Multimodal Models (LMMs) to deeply integrate image interaction with long-horizon reasoning capabilities remains a long-standing challenge in this field. Recent advances in vision-centric reasoning explore a promising "Thinking with Images" paradigm for LMMs, marking a shift from image-assisted reasoning to image-interactive thinking. While this milestone enables models to focus on fine-grained image regions, progress remains constrained by limited visual tool spaces and task-specific workflow designs. To bridge this gap, we present V-Thinker, a general-purpose multimodal reasoning assistant that enables interactive, vision-centric thinking through end-to-end reinforcement learning. V-Thinker comprises two key components: (1) a Data Evolution Flywheel that automatically synthesizes, evolves, and verifies interactive reasoning datasets across three dimensions-diversity, quality, and difficulty; and (2) a Visual Progressive Training Curriculum that first aligns perception via point-level supervision, then integrates interactive reasoning through a two-stage reinforcement learning framework. Furthermore, we introduce VTBench, an expert-verified benchmark targeting vision-centric interactive reasoning tasks. Extensive experiments demonstrate that V-Thinker consistently outperforms strong LMM-based baselines in both general and interactive reasoning scenarios, providing valuable insights for advancing image-interactive reasoning applications.

Figures (12)

• Figure 1: The three paradigms of vision-centric reasoning.
• Figure 2: Representative examples of V-Thinker's knowledge-driven synthesis spanning diverse reasoning domains.
• Figure 3: The rendering process from code to image.
• Figure 4: The Data Evolution Flywheel framework: Left: knowledge-driven evolution mechanism. Middle: coordinated calibration and progressive expansion stages. Right: representative synthetic QA instances generated through the flywheel.
• Figure 5: A representative sample from the synthesized dataset V-Interaction-400K ($\mathcal{D}$).