Gemini 3: Launch day reactions

In this episode of Release Notes, Logan Kilpatrick interviews Tulsi Doshi and Josh Woodward from Google DeepMind about the new Gemini 3 model. They discuss the model's state-of-the-art reasoning, multimodal understanding, and coding capabilities, as well as its well-roundedness and usability. The model is being released across various platforms, including the Gemini app, AI Studio, and Vertex, with external partners also shipping the model. The conversation explores the collaboration between DeepMind and product teams, the feedback loops, and the trade-offs in model development. They share anecdotes about the model's performance in vibe coding, multilingual abilities, and video understanding. The discussion also touches on the challenges of compute demand and the allocation of resources across different use cases, as well as the future development of smaller Gemini 3 models.