Vision-DeepResearch: Incentivizing DeepResearch Capability in Multimodal Large Language Models

TL;DR

This paper introduces Vision-DeepResearch, a multimodal large language model capable of multi-turn, multi-entity, multi-scale visual and textual search, significantly improving deep-research capabilities in noisy real-world scenarios.

Contribution

It presents a novel multimodal deep-research paradigm with extensive reasoning steps, training methods, and robust search strategies, outperforming existing models and workflows.

Findings

Supports dozens of reasoning steps and hundreds of engine interactions.

Outperforms existing multimodal deep-research models and workflows.

Achieves robust performance under heavy visual noise.

Abstract

Multimodal large language models (MLLMs) have achieved remarkable success across a broad range of vision tasks. However, constrained by the capacity of their internal world knowledge, prior work has proposed augmenting MLLMs by ``reasoning-then-tool-call'' for visual and textual search engines to obtain substantial gains on tasks requiring extensive factual information. However, these approaches typically define multimodal search in a naive setting, assuming that a single full-level or entity-level image query and few text query suffices to retrieve the key evidence needed to answer the question, which is unrealistic in real-world scenarios with substantial visual noise. Moreover, they are often limited in the reasoning depth and search breadth, making it difficult to solve complex questions that require aggregating evidence from diverse visual and textual sources. Building on this, we propose Vision-DeepResearch, which proposes one new multimodal deep-research paradigm, i.e., performs multi-turn, multi-entity and multi-scale visual and textual search to robustly hit real-world search engines under heavy noise. Our Vision-DeepResearch supports dozens of reasoning steps and hundreds of engine interactions, while internalizing deep-research capabilities into the MLLM via cold-start supervision and RL training, resulting in a strong end-to-end multimodal deep-research MLLM. It substantially outperforming existing multimodal deep-research MLLMs, and workflows built on strong closed-source foundation model such as GPT-5, Gemini-2.5-pro and Claude-4-Sonnet. The code will be released in https://github.com/Osilly/Vision-DeepResearch.