Skywork-R1V4: Toward Agentic Multimodal Intelligence through Interleaved Thinking with Images and DeepResearch
This addresses the need for more integrated and efficient multimodal AI agents, offering a novel approach that avoids costly reinforcement learning, though it appears incremental in combining existing capabilities.
The paper tackled the problem of multimodal agentic systems by developing Skywork-R1V4, a 30B parameter model that unifies multimodal planning, image manipulation, and deep search, achieving state-of-the-art scores of 66.1 on MMSearch and 67.2 on FVQA, surpassing Gemini 2.5 Flash on all 11 metrics.
Despite recent progress in multimodal agentic systems, existing approaches often treat image manipulation and web search as disjoint capabilities, rely heavily on costly reinforcement learning, and lack planning grounded in real tool-execution traces. To address these limitations, we present Skywork-R1V4, a 30B (A3B) parameter multimodal agentic model that unifies multimodal planning, active image manipulation ("thinking with images"), deep multimodal search, and, most critically, interleaved reasoning that dynamically alternates between visual operations and external knowledge retrieval. Trained solely via supervised fine-tuning on fewer than 30,000 high-quality, planning-execution-consistent trajectories and validated through stepwise consistency filtering, Skywork-R1V4 achieves state-of-the-art results across perception and multimodal search benchmarks: it scores 66.1 on MMSearch and 67.2 on FVQA, surpassing Gemini 2.5 Flash on all 11 metrics. Skywork-R1V4 exhibits emergent long-horizon reasoning at inference time, successfully orchestrating more than 10 tool calls to solve complex, multi-step tasks. Our results demonstrate that sophisticated agentic multimodal intelligence can be achieved through carefully curated supervised learning alone, without any reliance on reinforcement learning.