Researchers have developed FinAcumen, a framework for financial multimodal reasoning that uses selective experience memory to improve tool-augmented agents. This approach allows the agent to accumulate and reuse financially grounded reasoning experiences from prior trajectories, reducing unreliable tool routing, noisy retrieval, and hallucination-prone reasoning. FinAcumen was tested on four benchmarks and showed consistent improvements over finance-specialized models and proprietary general-purpose models. --- Why it matters: This matters because it addresses a key challenge in financial multimodal reasoning: the ability to coordinate numerical computation, visual interpretation, and temporal grounding across heterogeneous evidence sources. By improving tool-augmented agents' reliability and reducing errors, FinAcumen can help make more accurate predictions and decisions in high-stakes financial settings. Source: https://arxiv.org/abs/2606.17642