On November 19, Google unveiled its newest AI model, Gemini 3. DeepMind's CEO, Demis Hassabis, along with the head of the Gemini team, Josh Woodward, granted an exclusive interview to delve into the model's pivotal advancements. In this iteration, Gemini 3 has, for the inaugural time, embraced Generative UI capabilities, empowering it to dynamically construct interactive applications tailored to user requirements. For instance, a query about Van Gogh's life instantly spawns a comprehensive learning page replete with images and a timeline. Conversely, intricate calculation demands can generate bespoke mortgage calculators, marking a transition from merely 'answering queries' to 'crafting experiences.'

The model's reasoning prowess has witnessed a substantial boost. Earlier models often 'lost their way' during 5-6 step convoluted logical deductions. In contrast, Gemini 3 can seamlessly execute 10-15 steps of coherent reasoning, facilitating intricate tax planning, international travel itineraries, and extensive code system debugging. In the interdisciplinary PhD-level problem set dubbed 'Human Ultimate Exam,' Gemini 3 Pro's score soared from 21.6% to 37.5%, significantly outpacing GPT-5.1's 26.5%. Its SimpleQA test accuracy soared to 72.1%, more than doubling GPT-5.1's performance and markedly curbing hallucination occurrences.

In the realm of visual intelligence, the model has undergone a groundbreaking upgrade, securing a 72.7% score in the ScreenSpot-Pro screen comprehension test, a staggering 20-fold increase over GPT-5.1's performance, paving the way for AI Agents to automate computer operations seamlessly.

When it comes to coding capabilities, the model supports 'atmospheric coding,' enabling it to generate fully functional and visually appealing user interface code based on natural language cues. Coupled with the new agent development platform, 'Google Antigravity,' users can articulate their needs in natural language, and the model will autonomously invoke tools, craft interfaces, and rectify bugs.

Google has unequivocally shunned the emotional companionship domain, positioning Gemini 3 as a supercharged productivity tool deeply intertwined with Google's ecosystem offerings. For instance, the model can comprehend the context of a user's email, automatically categorize messages, and draft replies, tailoring the tone and content to align with the user's unique style.