New Models and Multimodal Capabilities
Google released Gemini 3.5 Flash, the new default model in AI Studio. It costs about 1.5 cents to analyze a 5-minute video and output a structured table. The 3.1 Flashlite model cuts latency further. Nano Banana for video (Omni Flash) generates and edits video from any input. The live API enables real-time voice conversations. A sample demo showed a dinosaur video – 31,000 tokens consumed, facts returned in seconds. Grounding with Google Search adds citations.
AI Studio: From Prompt to App
Build mode generates native Android apps from a text prompt. Ammaar Reshi described a foldable piano app: the top half shows a tutorial, the bottom half plays keys. The app is built in Kotlin, installed via USB or the Android emulator. Play Store distribution is coming. Workspace integrations – Gmail, Calendar – are rolling out today. A demo showed a “calendar roulette” app that deletes meetings, using OAuth and Firebase for authentication.
Open Models and On-Device AI
Gemma 4 supports 140 languages with a 256,000-token context window. It runs on laptops and Pixel 10 for on-device agentic use cases. A nonzero number of Gemma API calls are free inside Gemini APIs. Offline usage works – Paige Bailey described an 8-hour plane ride with Gemma 4 for vibe coding. The Google AI Edge gallery helps reduce project costs. Community partners include Unsloth and Kaggle.
Robotics and World Models
Gemini Robotics 1.6 runs on commodity hardware. A Stanford robot dog called Pupper uses the Gemini Live API with no training required – it follows a speaker and fetches balls. Genie 3, the world model family, generates interactive 3D scenes from text prompts. A demo showed a paper airplane navigating a waterfall with realistic physics and light reflections. Genie 3 is available with a Google AI Ultra subscription.
Notable Quotes
you won’t have to deal with any of the setup of Android apps or getting a simulator or anything like that Ammaar Reshi · ▶ Watch (12:30)
we’re building kind of the first generation of tools for people who are AI native Paige Bailey · ▶ Watch (28:36)
I keep telling people all you literally have to do is just go to that prompt box and describe your idea Ammaar Reshi · ▶ Watch (28:52)
It’s the first dog that you do not have to train Paige Bailey · ▶ Watch (26:47)
Key Takeaways
- Gemini 3.5 Flash costs ~1.5 cents per 5-minute video analysis in AI Studio.
- AI Studio Build generates native Android apps from a single text prompt.
- Gemma 4 runs on-device with 256K context and 140 languages.