Google for Developers Video Summaries

Gemini 3.5 Transcribe, a new LLM-based transcription model, is now available on both the Interactions API and the Live API. It demonstrates improved accuracy for alphanumerics, email addresses, phone numbers, units, and multilingual speech.

■ Model overview and availability
- Available on Interactions API and Live API
- LLM-based approach for accurate transcription

■ Customization features
- Custom vocabulary for names and terms
- Language hints to improve accuracy

■ Demonstrated capabilities
- Email addresses, phone numbers, and unit conversions
- Automatic recognition of 70+ languages

Developers interested in real-time or high-accuracy transcription will learn about the model's features and how to apply them in their own applications.

This video demonstrates a live translation broadcast application built with the Gemini 3.5 Live Translate model, LiveKit, and Google Cloud Run. It explains how the demo works, how to set it up locally, and how to deploy it to production.

■ Demo Overview
- Live translation broadcast with event IDs and language selection
- Session management: one session per target language, reusing existing sessions
- Use cases for live events and presentations

■ Technical Implementation
- Using Next.js, LiveKit, and the Gemini API
- WebSocket connections and WebRTC for audio and captions
- Translation bridge and session manager code

■ Deployment and Scaling
- Deploying to Google Cloud Run with Docker and Secret Manager
- Scaling limitations: max one instance, 15–20 simultaneous languages
- Recommendations for scaling beyond the demo

This video is for developers interested in integrating live translation into their own applications. Viewers will learn how to build and deploy a similar system and understand the architectural considerations involved.

We explore how enterprises balance frontier AI capabilities with cost efficiency, focusing on Gemini Flash 3.7's role in powering scalable agent workflows. The discussion covers model selection, multi-agent patterns, and practical strategies for reducing token costs while maintaining accuracy.

- Enterprise AI challenges: processing thousands of documents and making millions of data-grounded decisions per day.
- Cost optimization: using efficient models like Gemini Flash 3.7 to lower token costs and latency without sacrificing accuracy.
- Multi-agent systems: main agents dispatching sub-agents for parallel exploration, with low-latency models handling intent classification and planning.
- Developer impact: cheaper agents enable more experimentation and a broader range of applications.

This discussion is for developers, enterprise architects, and AI practitioners looking to deploy cost-effective agent systems. Viewers will gain insights into model tiering, agent orchestration, and how to maximize budget for AI initiatives.

Gemini 3.7 Flash is introduced as a workhorse model for coding and agents, with a focus on game development. The demonstration covers building a single-level animated sprite-based 90s game, using Nano Banana for all visual assets. It also shows how the model adapts to a completely new prompt and remixes the game into a different concept.

- Building a 90s sprite game: asset generation, an expansive map, multiple opponents, and interactive details such as flame reflections.
- Remixing with a new prompt: changing the theme from sorcerers to a pizza delivery driver in a suburban neighborhood, resulting in “Pizza Rush” with interactions like feeding pizza to dogs and raccoons.
- Improvements and availability: enhanced debugging, web development, and design adherence; rolling out in Google Antigravity, AI Studio, and the Gemini API.

Developers and creators interested in AI-assisted game development and rapid prototyping will find practical examples of prompt-based iteration. Viewers can learn how Gemini 3.7 Flash generates assets, follows design prompts, and transforms a game concept efficiently.

At the YC Startup School afterparty, founders, engineers, and students discuss how they integrate Google's AI models, including Gemini, Gemma, and AlphaFold, into their daily workflows. The conversation covers real-world applications ranging from document parsing to scientific research, offering practical insights for anyone exploring AI tools.

■ AI in Document Processing
- Using Gemini Flash to parse receipts and multilingual travel documents.
- Benefiting from Gemini's large context window and fast processing speed.

■ AI as a Development Partner
- Founders relying on Gemini and Gemma as a co-founder across various stages of their startup journey.

■ AI in Scientific Research
- Applying AlphaFold to visualize bacterial mutations and study antibiotic resistance.
- Using Gemini's research mode to synthesize neuroscience ideas and literature.

■ Model Selection and Offline Use
- Downloading Gemma for offline study during flights.
- Balancing model complexity and speed, with Gemini Flash praised for efficiency.

This recording suits founders, engineers, and students interested in practical AI applications. Viewers will learn how AI tools can accelerate business and research tasks, along with considerations for choosing the right model for specific needs.

This video provides an in-depth look at Google DeepMind's Gemini Robotics 2 suite, a set of AI models designed to give robots embodied reasoning, whole-body control, dexterity, and multi-robot collaboration. The researchers discuss the technical journey, current challenges, and how the models are being made available.

Topics covered include:
- The three key capabilities of Gemini Robotics 2: whole-body intelligence, dexterity, and multi-robot collaboration
- The research history: from reinforcement learning and transformers to vision-language-action (VLA) models, leading to the first Gemini Robotics release
- The unsolved problem of dexterous manipulation, the importance of diverse data, and the limitations of current sensing
- Deployment perspectives: industrial environments as near-term targets, home robots as a longer-term goal, and an estimated five-to-ten-year timeline for general-purpose robots
- Availability: Embodied Reasoning 2.0 via AI Studio and the Gemini Enterprise Agents Platform, plus a trusted tester program for on-device action models

This video is intended for those interested in robotics, embodied AI, and Google DeepMind's latest work. Viewers will gain a clear understanding of the Gemini Robotics 2 model suite, the technical obstacles that remain, and the practical roadmap toward real-world deployment.

This tutorial demonstrates how to set up tracing for a voice agent built with the Google ADK and the Gemini Live model using LangSmith. It walks through enabling the LangSmith ADK plugin, registering it on the runner, and capturing both user and agent audio to generate a complete trace.

■ Setting Up Tracing
- Registering the LangSmith ADK plugin
- Configuring project name, tags, metadata
- Capturing user audio and agent audio

■ Exploring the Trace
- Viewing audio replay and conversation transcript
- Inspecting tool calls and event details
- Checking token usage and cost breakdown

Developers building voice agents with Google ADK and Gemini Live can add observability and testing to their production workflow. By the end, viewers will know how to trace voice agent interactions to LangSmith and use the data for evals, datasets, dashboards, and debugging.

This episode covers the stable release of Android Studio Quail 2, updates to the Android Bench model evaluation leaderboard, a new series on AI prototyping at YouTube, and the launch of the Build with Gemini XPRIZE competition.

■ Android Studio Quail 2
- Parallel multi-task agent conversations, LeakCanary integration in Profiler, and App Quality Insights with agent mode

■ Android Bench and Emergent
- Updated leaderboard with new models and the Harbor framework for standardized evaluation; the new Emergent series reveals YouTube's AI prototyping stack

■ Build with Gemini XPRIZE
- $2 million total prize across five categories, with prototype submissions due by August 17

Viewers interested in Android development tools, AI model evaluation, and AI-driven business opportunities will gain practical updates and links to relevant resources.