Medical workflows often require several interdependent analyses, yet existing specialist and generalist models are not built for coordination. We introduce MMedAgent-Plus, a multimodal medical tool-use agent. It couples a vision‑language planner with 18 leading specialist models covering 12 tasks and 8 modalities. Through a unified tool interface, the planner selects appropriate specialists, executes them sequentially, and synthesizes a final answer grounded in their outputs. The planner is instruction‑tuned on 208,928 samples from 32 public medical benchmarks, encompassing single‑ and multi‑tool interactions across single and multiple rounds. We evaluate tool orchestration, response synthesis, and end‑to‑end execution separately. Results show tool‑call accuracies of 91.0% in single‑tool multi‑round, 98.1% in multi‑tool multi‑round, and 80.7% in multi‑tool single‑round settings. Given correct tool outputs, MMedAgent-Plus attains 14.4‑14.8 out of 15 for faithfulness, completeness, and conciseness on 600 evaluation samples. In end‑to‑end tests without pre‑supplied tool outputs, it correctly invokes the required specialist in 90.6% of 350 cases spanning seven tasks. Final‑answer accuracy reaches 0.98, matching the top specialist and far exceeding the strongest baseline (0.28).
Review