Overview
- Google has activated Gemini 3 in Search’s AI Mode for U.S. Google AI Pro and Ultra subscribers, adding a Thinking setting, upcoming automatic model selection, and new generative UI that can create interactive tools and simulations such as calculators or guided explainers.
- The flagship Gemini 3 Pro accepts text, images, video, audio and PDFs with up to a 1,048,576‑token context and structured JSON output, and it posts leading scores including 1501 Elo on LMArena and strong results on GPQA Diamond and MMMU benchmarks.
- Developer and enterprise access spans AI Studio, the new Antigravity platform, Vertex AI, Gemini Code Assist and the Gemini CLI, enabling agentic, multi‑step coding and workflow execution beyond inline completions.
- Hands-on testing found the model can generate interactive 3D visuals and magazine-style layouts and perform agent tasks like surfacing key emails and adding reminders, yet outputs often fell short of demo polish and agents were slow or error‑prone on actions like reservations.
- Influencers showcased one‑shot builds such as a 3D LEGO editor and game recreations, even as users flagged instruction adherence issues and pricing complaints, and Google’s higher‑intensity Deep Think remains gated for premium tiers pending safety reviews; a TechCrunch anecdote underscored tooling limits when the model lacked live search.