Overview
- Gemini 3 Pro is now live across Search, the Gemini app, AI Studio, Vertex AI, the Gemini CLI and the Antigravity IDE, with the Deep Think mode limited to premium and Ultra users and testers pending safety checks.
- The model accepts text, images, video, audio and PDFs, supports about 1,048,576 tokens of context, and caps outputs at 65,536 tokens.
- Google cites benchmark leadership, including a 1501 Elo on LMArena and strong scores on Humanity’s Last Exam, GPQA Diamond and MathArena Apex.
- Hands-on tests find useful generative interfaces and interactive visualizations in Canvas and new “visual” layouts, though results often look rougher than Google’s demos.
- Trials of agentic features show helpful automations like inbox triage and task creation, yet reviewers report slow or inconsistent transactional behavior and hallucinations that can produce misleading, sometimes defensive responses.