Google AI

AI Studio's vibe coding mode now ships backends: Cloud Firestore databases, Firebase authentication, securely stored API credentials for outside services, and multiplayer apps. The Antigravity agent keeps project structure and chat history, and continues multi-step edits while you are away.

Google Labs rebuilt Stitch around a context-aware design agent that accepts text, image, or code as creative input, critiques builds, and drafts PRDs while emitting production-ready front-end code. A DESIGN.md file carries design rules between Stitch and other coding tools.

Google is opening Project Genie, a web prototype that turns a prompt into an explorable interactive world, to AI Ultra subscribers in the US. It combines Genie 3's simulated physics and interaction with Nano Banana for sketching and design and Gemini for world knowledge and reasoning.

Google's new audio model performs streaming speech-to-speech translation across more than 70 languages, starting to translate while the speaker is still talking and holding pacing, pitch and intonation over extended sessions. Live now in Google Translate on iOS and Android.

DeepMind's Lyria 3 turns text prompts or photographs into music tracks, with deliberately high prompt character limits for detailed direction and Gemini-suggested lyrics that fit a stated theme. Rolling out in beta in the Gemini app for users 18 and over.

Gemini 3.6 Flash answers efficiency complaints about 3.5 Flash with better coding and multimodal work using substantially fewer tokens per task. Gemini 3.5 Flash-Lite is the fast, cheap tier at roughly 350 output tokens per second. Both ship in the Gemini API and AI Studio.

Google's open models arrive as 31B dense and 26B MoE variants plus E2B and E4B edge builds with native vision and audio. Apache 2.0 licensing, function calling with structured JSON output, 256K context, and support for 140+ languages.

Google's new low-cost tier is priced at $0.25 per million input tokens and $1.50 per million output, with roughly 45 percent faster output than Gemini 2.5 Flash and selectable thinking levels so you can dial reasoning per task. In preview via the Gemini API in AI Studio and Vertex AI.

Google's Omni model takes text, audio, image, and video inputs and blends them into consistent video, then accepts natural language edits to change characters, settings, or style. Omni Flash is live for AI Plus, Pro, and Ultra subscribers, and free in YouTube Shorts and YouTube Create.

Google's text-to-speech model takes bracketed style cues embedded directly in the script to control tone, pace, and delivery. It covers 70+ languages with 24 evaluated at high quality, and ships in preview through the Gemini API and AI Studio.

Google's newest workhorse model pushes on multi-step planning and tool calls, with effort controls that scale thinking tokens to task difficulty. It is in the Gemini API via AI Studio and Android Studio, in Antigravity for agent workflows, and in Gemini Enterprise.

A prompting guide for Gemini Omni video generation: lean on the model's world knowledge rather than over-describing scenes, control typography and text placement, direct shots with cinematography terms, make targeted edits without rewriting the prompt, and alter subject action mid-scene.

Google DeepMind's newest weather model skips physics simulation and reanalysis pipelines. It takes live geostationary satellite feeds as input, trains on direct surface and atmospheric observations, refreshes a global forecast every hour, and claims roughly 5x the spatial sharpness of WeatherNext 2.

The updated video generation and editing model carries over Veo's creative controls and adds 4K upscaling, first and last frame control and fast 360p drafting. Scene extension now conditions on 10 seconds of the original video instead of one second. Available in Flow, the Gemini app and AI Studio.

Google's realtime voice model ships with faster responses, a context window twice as long inside Gemini Live, and coverage across 200 more regions. Developers get it through the Live API and AI Studio in preview, alongside rollout in the Gemini app, Search Live, and Gemini Enterprise.

Google AI Pro and Ultra subscribers now get raised usage limits and access to Nano Banana Pro and Gemini Pro models inside Google AI Studio by signing in with the subscriber account, with no API key required.

Veo's Ingredients to Video mode gains vertical formats with steadier characters, objects, and backgrounds. Upscaling to 1080p and 4K arrives across all Veo models, and the Gemini app can now verify whether a video was generated by Google AI.

TPU 8t targets training with roughly 3x the compute of the prior generation, 10x faster data delivery, and 97% goodput under hardware failure. TPU 8i targets agent serving with triple the on-chip memory, 80% better performance per dollar, and 5x lower latency.

Google details the product-specific engineering behind Gemini in Workspace: teaching the model the native Docs data model rather than Markdown so formatting and styles survive, a custom translation layer that keeps generated Slides editable, and a 70.48% result on the full SpreadsheetBench dataset.

Gemini 3 Deep Think's update pushes it from abstract reasoning toward applied engineering, with reported use optimizing crystal growth for high-temperature semiconductor candidates and iterating physical prototypes. It is available in the Gemini app for AI Ultra subscribers, with benchmarks publishe

Google's most cost-efficient Gemini 3 model enters preview through the Gemini API in AI Studio and Vertex AI. The demos aim at volume work that was previously too slow or expensive: bulk image sorting and classification, plus a retail agent running multi-step reporting and dashboard automation.

Google opens the Gemini 3.5 series with 3.5 Flash, positioned on agentic and coding performance rather than general chat. The post traces the lineage from understanding in Gemini 1, reasoning in Gemini 2, and generation in Gemini 3, and marks 3.5 as the agent-oriented step.

Google's music model now understands song architecture, so prompts can specify intros, verses, choruses, and bridges, and tracks run to three minutes instead of thirty seconds. Access covers AI Studio and the Gemini API as well as Vertex AI, the Gemini app, and Google Vids.

Google's prompting guidance for Gemini 3.1 TTS: bracketed inline tags such as [whispers] or [slow], placed exactly where delivery should change and never adjacent to each other, mixing emotion, pacing, pause length, and vocalization cues in a single script.