Google recently unveiled that its latest flagship large model, Gemini 4 Argon, is now accessible to developers and enterprises globally. Paid users can also leverage this model within the Gemini app and in Search AI mode. This model has showcased outstanding performance across numerous benchmark tests, outperforming rival models from OpenAI and Anthropic on the Vals AI Composite Index. Leveraging Gemini 4 Argon, researchers were able to organize data, fine-tune models, and establish screening rules within a matter of hours, successfully designing five innovative artificial serine esterases. Laboratory tests confirmed that these designs boasted robust catalytic activity, suggesting significant potential applications in pharmaceutical manufacturing and fine chemical engineering. In contrast to the serine hydrolases previously crafted by Nobel laureate David Baker’s team using specialized models such as RFdiffusion, Gemini 4 Argon, as a versatile large model, can intervene in multiple facets of the scientific research process. This includes data processing, model adjustment, and the formulation of screening strategies, thereby substantially expediting specialized research workflows. At present, Google has not fully disclosed the complete toolchain utilized for this enzyme design, and the performance advantages of Gemini 4 Argon necessitate further experimental validation. The role of versatile large models is evolving from merely answering scientific research queries to organizing actual research processes. These models can synergize with specialized models like AlphaFold, where professional models tackle specific computational tasks while versatile models manage procedural work. This collaboration reduces the barrier to using scientific research tools. Moreover, Google intends to offer up to 10,000 eligible scientists complimentary access to the highest usage tier of Google Antigravity for six months, aiming to facilitate the model’s integration into more real-world scientific research projects.