Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search

2025-09-09

AI Summary

Existing open-source multimodal models suffer from limited reasoning paradigms and restricted interaction rounds (typically ≤3), hindering deep, goal-persistent visual search requiring iterative trial-and-error. This paper introduces the first scalable multi-round reasoning framework tailored for visual search. Methodologically, it (1) constructs a novel exploratory visual search dataset; (2) designs diverse cold-start trajectory generation and cross-turn masking to explicitly model long-range dependencies; and (3) integrates image-tool invocation, reinforcement learning–driven iterative data collection, and policy optimization. The resulting model supports数十 rounds of natural, grounded interaction. Evaluated on multiple challenging visual search benchmarks, it achieves state-of-the-art accuracy, with performance consistently improving across interaction rounds. Crucially, it significantly enhances reasoning path diversity and chain-of-thought depth—demonstrating robust, adaptive, and interpretable visual search capabilities.

Technology Category

• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Scaling up tool-based interactions for complex visual search
• Enabling multi-turn reasoning with diverse exploration patterns
• Overcoming limited interaction turns in multimodal models

Innovation

• Scaling up tool-based interactions for deep multi-turn reasoning
• Iterative data collection pipeline for diverse reasoning patterns
• Over-turn masking strategy to balance training efficiency and scalability