Overview
- Google shipped Gemini 3 on November 18 and embedded it in Search, the Gemini app, Vertex AI, AI Studio and enterprise offerings used by hundreds of millions.
- Google cites top scores across public benchmarks, including LMArena and GPQA Diamond, while some experts caution that such tests can overstate real-world reliability.
- Early hands-on demos highlight strong multimodal performance for coding, UI design, 3D and physics simulations, and video analysis, including a widely shared Matthew Berman showcase.
- Google’s higher-reasoning Deep Think mode remains safety-gated, with broader availability planned after additional evaluations for Google AI Ultra subscribers.
- An internal memo from Sam Altman flagged “temporary economic headwinds” for OpenAI, as CFO Sarah Friar reported softer ChatGPT engagement but reiterated the company’s 2025 revenue target.