Benchmarking multimodal large language models for medicinal plant identification

With the rapid advancement of artificial intelligence technology, large language models (LLMs) have shown considerable potential in the medical field. This study systematically evaluated the performance of five multimodal LLMs, including GPT-4o, Llama4Scout, Gemma3-27B, Qwen3-VL235B, and DeepSeek-VL2, in medicinal plant image recognition tasks. We selected 200 medicinal plant images from a privately copyrighted dataset and tested the models using a four-choice format to evaluate their recognition accuracy. Results indicate that Qwen3-VL-235B demonstrated the highest accuracy at 90.50%, outperforming other models. GPT-4o ranked second at 85.00%, while Gemma3-27B, Llama4Scout, and DeepSeek-VL2 achieved 65.50%, 58.50%, and 58.00%, respectively. These findings indicate that multimodal LLMs hold promising potential for medicinal plant recognition, yet further optimization for specific domains is required to enhance accuracy.