Overview
- Gemini 3 is now live across Google surfaces, with AI mode in Search limited to Google AI Pro and Ultra subscribers in the U.S. and a new Thinking setting for tougher queries.
- Google is adding generative UI in Search responses, producing interactive layouts and tools such as simulations or calculators when relevant.
- Deep Think, the higher-intensity reasoning mode, remains gated for safety reviews and will reach premium tiers first.
- Model documentation cites top benchmark results and a 1,048,576‑token multimodal context window with up to 65,536 tokens of output for API use.
- Hands-on reports show impressive one‑shot code+UI generation and agentic workflows, alongside uneven visual fidelity, occasional execution errors, reliance on connected tools for fresh facts, and influencer concerns about token pricing.