Multimodal large language models (MLLMs) have made substantial strides in chart understanding and reasoning, yet their performance varies dramatically across chart types. Modern MLLMs excel on common charts but see a sharp drop in accuracy on long‑tail charts that feature diverse visual structures and complex reasoning demands. Tool‑augmented agents can boost complex chart reasoning by adding specialized visual capabilities, but manually engineering strategies and tools for the heterogeneous and evolving long tail is hard to scale. To address this, we introduce ChartEvo, a recursive self‑improvement framework that enables chart agents to adapt to underperforming chart types through experience‑driven harness‑level self‑improvement. ChartEvo jointly evolves reusable chart‑specific Skills and executable Tools, while learning reusable local tool‑call chains to compose the evolved capabilities and control the inference overhead of tool‑augmented reasoning. Experiments on long‑tail charts from ChartBench and ChartX show that ChartEvo improves accuracy over its base model Qwen3.7‑Plus by 45.6% and 32.5%, respectively, demonstrating the effectiveness of recursive self‑improvement for long‑tail chart adaptation.
Review: ChartEvo illustrates how self‑evolving agents and modular tool chains can achieve adaptive performance on diverse chart formats, offering a promising direction for enhancing MLLM robustness in real‑world applications.