Our DARWIN LLM has achieved a groundbreaking milestone, having been recognized as the world’s most accurate model in Experimental Bandgap Prediction and Metal Classification Model within MatBench. This marks the first instance where a fine-tuned Large Language Model has surpassed all existing Machine Learning models in Material Design tasks. Impressively, our model has also outperformed other fine-tuned GPT models. This achievement sets a new leaderboard in the field of Material Science and AI. Details of the benchmark could check below: https://matbench.materialsproject.org/Leaderboards%20Per-Task/matbench_v0.1_matbench_expt_gap/