Paper 2511.04460
V-Thinker: Interactive Thinking with Images
- Published
- Nov 2025
- Research lab
- Independent
- Citations
- 19
- GitHub
- 177 stars
01 In brief
Summary
V-Thinker is a general-purpose multimodal reasoning assistant that enables interactive, vision-centric thinking through end-to-end reinforcement learning.
It comprises a Data Evolution Flywheel that automatically synthesizes, evolves, and verifies interactive reasoning datasets across diversity, quality, and difficulty, producing the V-Interaction-400K dataset.
A Visual Progressive Training Curriculum first aligns perception via point-level supervision using V-Perception-40K, then integrates interactive reasoning through a two-stage framework with supervised fine-tuning and reinforcement learning.
The authors introduce VTBench, an expert-verified benchmark with 1,500 QA pairs across perception, instruction-guided interaction, and interactive reasoning tasks.
Experiments show V-Thinker-7B outperforms strong baselines like GPT-4o and Qwen2.5-VL-7B on VTBench, with an average accuracy improvement of 12.5%, and also improves general reasoning benchmarks (MathVision +6.3%).
Ablations confirm all training stages are effective, with RL being essential.
The Data Evolution Flywheel scales knowledge concepts and visual tools non-linearly, reaching 24,767 nodes across 25 domains after five iterations.
02 From the paper
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.