ActICL: Post-Training for Agentic Multimodal In-Context Learning

Multimodal in-context learning (ICL) is an important capability of multimodal large language models (MLLMs). Its performance is closely tied to the demonstrations it uses. Thus, retrieving demonstrations from a given bank is a core problem in many ICL-based systems. Common methods use fixed metrics to retrieve all the demonstrations at once, which may introduce redundancy and noise. Context-aware retrieval methods have emerged to address this problem. However, none of them lets the MLLM itself learn how to judge what new demonstrations it needs through its reasoning and then carry out progressive retrieval on its own. To this end, we introduce ActICL, a framework that trains an MLLM agent to acquire and use an ICL context via tool-augmented reasoning. ActICL builds an agent that analyzes its own information gaps during iterative reasoning and tool-use actions. Based on these gaps, it performs targeted retrieval to acquire a set of initial candidates. It can then use the existing context to further filter the candidates and refine the context, or it can iteratively perform a new retrieval. When the agent judges that the context is sufficient for the task, or when the resource limit is reached with the user request met, it sends the acquired context to an answering model, either the agent itself or another MLLM. We use a two-stage post-training procedure that jointly optimizes the agent's acquisition and answering. When the agents adaptively decide the number of demonstrations, ActICL improves average performance by 11.6 percentage points over standard similarity-based retrieval across seven benchmarks and three MLLMs, and by 4.4 points compared to the strongest prior baseline. The contexts acquired by ActICL agents can also be used effectively by frontier models. When GPT-5.6-Sol performs 4-shot ICL, ActICL's contexts improve its performance by 3.5 points over standard retrieval, averaged over three benchmarks.