Paper 2509.07969
Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search
- Published
- Sep 2025
- Research lab
- Independent
- Citations
- 101
- GitHub
- 424 stars
01 In brief
Summary
Mini-o3 is a system for visual search that scales up tool-based interactions and reasoning patterns, achieving state-of-the-art performance on challenging tasks.
It addresses limitations of existing open-source VLMs, which show monotonous reasoning and limited interaction turns.
The approach includes three key components: constructing the Visual Probe Dataset with thousands of challenging visual search problems; developing an iterative pipeline to collect cold-start trajectories with diverse reasoning patterns (e.g., depth-first search, trial-and-error); and proposing an over-turn masking strategy in reinforcement learning to avoid penalizing responses that exceed the turn limit, enabling test-time scaling.
Despite training with only 6 interaction turns, Mini-o3 generates trajectories that scale to tens of turns at inference, with accuracy improving as turns increase.
Experiments show Mini-o3 outperforms baselines like DeepEyes and GPT-4o on VisualProbe, V* Bench, HR-Bench, and MME-Realworld, with significant gains on hard tasks.
02 From the paper
Abstract
Recent advances in large multimodal models have leveraged image-based tools with reinforcement learning to tackle visual problems. However, existing open-source approaches often exhibit monotonous reasoning patterns and allow only a limited number of interaction turns, making them inadequate for difficult tasks that require trial-and-error exploration. In this work, we address this limitation by scaling up tool-based interactions and introduce Mini-o3, a system that executes deep, multi-turn reasoning -- spanning tens of steps -- and achieves state-of-the-art performance on challenging visual search tasks. Our recipe for reproducing OpenAI o3-style behaviors comprises three key components. First, we construct the Visual Probe Dataset, a collection of thousands of challenging visual search problems designed for exploratory reasoning. Second, we develop an iterative data collection pipeline to obtain cold-start trajectories that exhibit diverse reasoning patterns, including depth-first search, trial-and-error, and goal maintenance. Third, we propose an over-turn masking strategy that prevents penalization of over-turn responses (those that hit the maximum number of turns) during reinforcement learning, thereby balancing training-time efficiency with test-time scalability. Despite training with an upper bound of only six interaction turns, our model generates trajectories that naturally scale to tens of turns at inference time, with accuracy improving as the number of turns increases. Extensive experiments demonstrate that Mini-o3 produces rich reasoning patterns and deep thinking paths, effectively solving challenging visual search problems.