I tested Gemini 3 and ChatGPT on three visual prompts—here’s what happened

Google's latest innovation, Gemini 3, has swiftly ascended to the forefront of AI performance rankings, capturing significant attention in the tech community following its impressive debut. Among its advocates, Salesforce CEO Marc Benioff praised Gemini 3 as an "extraordinary step forward," indicating that after just a brief interaction, he found it hard to return to using ChatGPT. To evaluate how these two AI models perform in practical applications, a series of tests were conducted, focusing on three distinct creative and visual tasks: infographic creation, image transformation, and geographical mapping. These evaluations provide valuable insights into how each tool addresses both straightforward and more intricate visual challenges. In the first task, participants were asked to generate an infographic titled "Climate-friendly habits for daily life," which required five sections with brief titles, descriptions, and a simple layout. Both Gemini 3 and ChatGPT succeeded in producing the infographic, but their methods varied markedly in terms of speed and visual quality. The second prompt involved transforming a user-uploaded image into a classic black-and-white studio portrait featuring a tuxedo and a vintage aesthetic, with specific lighting and backdrop requirements. This exercise revealed significant differences in how each model handled image processing and creativity. The final test, aimed at creating a detailed map of India, presented a challenge for both AI systems. Participants were tasked with labeling all states and union territories, while also providing a unique cultural, geographical, or economic fact for each area. Here, both models encountered difficulties, indicating their current limitations when faced with complex, factually dense graphics. Overall, while Gemini 3 excelled in terms of speed and visual refinement in the first two tasks, both AI models struggled with accuracy in the geography challenge. Ultimately, Gemini 3 outperformed ChatGPT in two out of the three assessments, although both still have room for growth in managing detailed factual visuals.