The latest AI news and updates of Google — the company behind Gemini, DeepMind research, Gemma open models, and AI across Search, Cloud, and Workspace. Covering Google's latest product updates, launches, and research from the past 90 days.
Gemini
Google Gemma
Google AI StudioGoogle Gemma17h agoSatlyt is running Google’s Gemma 3 1B model directly on satellites to analyze telemetry and diagnose faults locally. This deployment reduces diagnostic data payloads by over 64% by generating compact summaries before downlink. The company plans to launch Gemma 4 E2B on NVIDIA Jetson Orin Nano hardware next to add onboard reasoning and vision capabilities to its orbital infrastructure.
GoogleSep 11Google Research introduced ToolGrad, a framework that generates ground-truth tool-use chains before user prompts to streamline dataset creation. By reversing the traditional query-first paradigm, the system achieves a 99.8% annotation pass rate. This answer-first approach produces complex tool-use data at a lower cost, improving the tool-calling performance of fine-tuned models like Gemma-3.
Google LabsSep 10Google Labs released Dreambeans to all US users aged 18 and older on iOS and Android. The app generates personalized daily stories by synthesizing data from connected services like Gmail, Calendar, and YouTube. Gemini app integration adds chat context to the synthesis, while Nano Banana 2 generates personalized artwork for each story.
Google AI StudioSep 10Google AI Studio launched a preview of Gemini API documentation directly within its build environment. This integration consolidates resources, providing immediate access to API references while building. The classic documentation site remains available for users who prefer the original experience.
GeminiSep 10Google launched the Gemini app for Windows today, bringing AI assistance to Windows 10 and 11. The Alt + Space shortcut opens the interface over active work for drafting, task execution via Gemini Spark, and media generation. Gemini Spark and Gemini Omni capabilities require a Google AI subscription.
GoogleSep 10Google launched Google Pics, an image generation and editing tool built on its Nano Banana model. The tool supports object isolation, in-image text translation, and collaborative editing. Google Pics is available now to Google AI Pro and Ultra subscribers and most Workspace business customers, with native integration in Docs and Slides and upcoming support for Google Drive.
GoogleSep 9Google Research, in collaboration with HHMI Janelia, released the complete wiring diagram of a male fruit fly’s brain and central nervous system. This map documents over 166,000 neurons and 125 million connections, the largest proofread brain map to date. Having complete maps for both sexes allows researchers to pinpoint structural differences and study how they drive biological behaviors.
Google DeepMindSep 8Google DeepMind launched AlphaGenome Atlas, a searchable database mapping the predicted molecular effects of all 9 billion possible single-letter DNA changes in the human genome. The 1-petabyte dataset includes the AlphaGenome Variant Impact (AVI) score to rank mutation effects. Access is available via a web portal, API, and Google Antigravity, with commercial availability coming to Google Cloud.
AntigravitySep 8Google Antigravity integrated the AlphaGenome Atlas Skill into its scientific workbench to accelerate genomic discovery. AI agents use the skill to prioritize variants, generate structural plots, and build testable hypotheses. The Science Skills bundle, which integrates insights from AlphaGenome, AFDB, and UniProt, is available for download from the official GitHub repository.
Google AI StudioSep 4Google released Lyria 3.5, its latest music generation model, now available in AI Studio, the Gemini API, and the Gemini app. The model generates high-fidelity 44.1 kHz stereo audio, including full-length songs with verses and choruses, from text or image prompts. It supports custom lyrics, timestamp-based structural control, and includes SynthID watermarking for all generated audio.
Google DeepMindSep 3Google DeepMind launched WeatherNext 3, a global weather AI model that uses real-time satellite data to generate hourly, high-resolution forecasts. The model achieves 5km spatial resolution and improves precipitation accuracy by up to 50%. It is now integrated into Google Search, Gemini, and Maps, with data access available via BigQuery, Earth Engine, and Google Cloud.
Gemini NotebookSep 3Google is introducing flexible usage limits to Gemini Notebook, shifting limit resets from daily to every five hours. The update also adds deferred artifact generation, which queues supported outputs for later processing when capacity is exhausted. These changes provide more control over compute budgets and workflow management throughout the day.
Google GemmaSep 2Google now recommends Gemma 4 as the local model provider for Android Studio. This integration enables offline Agent Mode, ensuring code remains on the local machine without token quotas. The model handles multi-step tasks like feature design, refactoring, and build resolution, with hardware requirements ranging from 8GB to 24GB of RAM depending on the model variant.
GoogleSep 2Google launched Gemini 3.8 Flash Cyber, a specialized model for vulnerability detection and automated patch generation. The model achieves a 71.0% success rate on internal vulnerability discovery benchmarks across 20 programming languages. Google reports the model produced 2.6 times more correct patches for Chrome than other commercial models. Access is available to defenders through the Fairwind Program.
GoogleSep 2Google introduced Gemini 3.8 Flash, an intelligent workhorse model designed for long-horizon software engineering and autonomous agentic tasks. Alongside it, Google released Gemini 3.8 Flash Cyber, a specialized model for vulnerability detection and automated patching. Both models are available today, with 3.8 Flash priced at $0.75 per million input tokens and $3.75 per million output tokens.
GoogleSep 1Google Research released MAPL-EMIT, a deep-learning framework that automates the detection and quantification of methane plumes from space. The model uses a vision transformer architecture to analyze hyperspectral satellite data, achieving an 84% recall rate on expert-annotated plumes. Google Research also released a global plume database, synthetic training datasets, and an inference library to support climate mitigation efforts.
Google AI StudioSep 1Google launched agentic video understanding for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. This feature replaces static frame processing with an agentic loop that dynamically scans video, audio, and transcripts. It reduces analysis costs by up to 66% and token consumption by up to 88% while improving accuracy by 7% on standard benchmarks.
Google GemmaSep 1Google reports that its Gemma 4 26B A4B model now runs 130.3% faster on Apple Silicon hardware. This performance gain follows community-driven optimizations on the MLX Fast leaderboard, where developers refined the model's prefill and decode throughput. The update improves local inference efficiency for the 26-billion parameter model on Mac systems.
Gemini NotebookSep 1Google is expanding Gemini Notebook’s Short Video Overviews to support 70+ languages and three new English variants. The feature, which generates ~60-second vertical video summaries of source material, is rolling out on web and mobile for Ultra and Pro subscribers. The rollout is currently in progress for Pro users.
Google DeepMindSep 1Google DeepMind is bringing agentic video understanding to Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. This feature dynamically scans video segments, reducing token consumption by up to 88% and costs by up to 66% while improving accuracy by 7%. It is available via API in Google AI Studio and the Gemini Enterprise Agent Platform.
AntigravitySep 1Google Antigravity introduced /boost, a slash command activating an on-demand multi-agent reasoning pipeline for complex engineering tasks. The system decomposes challenges into subtasks, executes them via specialized agents, and verifies solutions across iterative rounds. Available in Antigravity 2.0 and CLI for Pro and Ultra subscribers, /boost consumes additional tokens to perform deep reasoning on difficult bugs and refactoring.
Google GemmaAug 31Google Gemma is highlighting Gilbeot, an on-device multimodal AI walking assistant for elderly users. The project runs locally on Android phones using the LiteRT-LM framework and custom speech LoRAs to deliver real-time, offline voice navigation. It translates complex map coordinates into natural language instructions based on visible objects, ensuring accessibility on older devices without requiring hardware upgrades.
GoogleAug 31Google Research released TimesFM-3, a 330-million parameter foundation model that enables zero-shot multivariate time series forecasting in a single forward pass. The model jointly predicts multiple related time series and incorporates external covariates, such as weather or promotions, to improve accuracy. It is now available on GitHub and Hugging Face, outperforming existing models on major public benchmarks.
Gemini NotebookAug 30Google launched Expert Intelligence, a cross-Google initiative adding eligible Google Play ebooks to Gemini Notebook as sources. The feature supports querying and synthesizing content from purchased books alongside other files. Future updates will expand support to third-party subscriptions, textbooks, and broader availability in the Gemini app and AI Mode in Search.
Google GemmaAug 28Google is running its Gemma 4 E2B model fully locally on an 8GB NVIDIA Jetson Orin Nano. The setup uses native image and audio input alongside function calling to perform vision, voice interaction, and camera control. This demonstration shows multimodal AI capabilities operating entirely within an 8GB memory footprint on power-constrained edge hardware.
Google AI StudioAug 28Google introduced Gemini Omni 1.1 Flash, a production-ready generative video model. The update adds scene extension with 10 seconds of prior context, first and last frame interpolation, and 4K upscaling. It also includes 360p draft generation for faster iteration and supports video references. The model is available via the Gemini API, Enterprise Agent Platform, and Google Flow.
AntigravityAug 28Google Antigravity researchers used the Teamwork multi-agent framework to solve seven open problems in theoretical computer science and mathematics, including Knuth’s Cycles Conjecture. The framework autonomously proposes, stress-tests, and builds solutions over hours or days. Teamwork also developed a cycle-accurate RISC-V CPU simulator and contributed performance optimizations to the Eigen and ParlayHash open-source libraries.
GoogleAug 27Google Research introduced the planetary prediction engine, an autonomous AI system that executes end-to-end geospatial modeling workflows from natural-language queries. By integrating statistical data with foundation model embeddings, the engine automates data discovery, feature engineering, and model optimization. This research capability reduces model-building time from weeks to minutes while improving predictive accuracy across public health and food security benchmarks.
Google DeepMindAug 27Google DeepMind is piloting double-blind evaluations for frontier AI models to prevent benchmark contamination. By using a secure cryptographic environment, the process ensures external evaluators cannot see model weights and Google cannot see test prompts. The pilot tests a Gemini Flash Lite model in partnership with organizations including the Singapore AI Safety Institute, OpenMined, AVERI, and MLCommons.
AntigravityAug 27Antigravity agents now create rich, interactive artifacts, including dynamic data visualizations and 3D simulations, rendered directly in the agent manager. These artifacts move beyond static markdown or diagrams, allowing direct interaction and inspection. The feature supports zero-dependency rendering and offline export to standalone HTML files, triggered via natural language prompts or the generative UI command.
GoogleAug 27Google Research introduced GlucoFM, a self-supervised foundation model for continuous glucose monitoring that separates slow glycemic trends from short-term deviations. The model achieves a 4.1-point absolute gain in PR-AUC over existing CGM-specific baselines across seven clinical prediction tasks, including diabetes risk assessment and insulin resistance. It uses a dual-stream architecture to process 109,066 hours of unlabeled sensor data.
GeminiAug 26Google updated Gemini Live to handle complex tasks hands-free, adding four productivity features. Daily Brief provides spoken morning digests, while Gemini Spark executes multi-step background projects. Personal Intelligence connects past conversations with Google apps, and voice-controlled Gmail management handles email triage. The update is available globally at no cost, with specific features requiring paid subscriptions.
GoogleAug 26Google introduced Gemini 3.5 Transcribe, a speech-to-text model that automatically removes filler words, formats unstructured speech, and executes voice commands using screen context. The model supports 85+ languages and is available in the Gemini macOS app, Gboard on Android, and via the Gemini API. Chrome and Gemini Enterprise for Customer Experience support arrive soon.
AntigravityAug 26Google Antigravity now integrates directly into Apple Xcode 27 beta 6 or later, bringing agentic coding to macOS, iOS, and iPadOS development. The integration performs autonomous project building and debugging within the editor, with onboarding managed through the Intelligence settings panel.
GoogleAug 25Google Research introduced AgentHands, an LLM-powered XR research prototype that augments conversational agents with synchronized, expressive hand gestures. The system maps verbal instructions to physical motions like pointing or tracing, providing spatially grounded guidance in 3D environments. By syncing gestures with speech, the prototype improves action understanding and reduces cognitive load during complex physical tasks in Android XR.
Google LabsAug 25Google Labs launched Play with Putty, a collaborative vibe coding experiment for building tools and websites in real time. The platform features instant sync for multiplayer editing, natural language instructions for UI components, and a shared history ledger for tracking project changes. Access is currently available via a US-based waitlist for individuals aged 18 and older.
GoogleAug 25Google released WeatherNext Cyclones, an AI model the U.S. National Hurricane Center used in real-time during the 2025 hurricane season. The model predicts cyclone track, intensity, and size simultaneously, providing an extra full day of warning. Google published the research in Nature and open-sourced the code and model weights on GitHub.
GoogleAug 25Google Research published the first episode of "Ask a Scientist," detailing how its Climate Crisis Resilience team uses AI to forecast riverine floods up to 7 days ahead and urban flash floods up to 24 hours in advance. The team also open-sourced its hydrology modeling framework on GitHub, allowing researchers and meteorological agencies to build on the technology.
GoogleAug 22Google Research introduced Mobility-Embedded POIs (ME-POIs), a framework that enriches text-based place representations with aggregated, anonymized mobility data. By blending static metadata with real-world activity rhythms, the model improves predictions for visit intent, price levels, and busyness. This approach creates holistic numerical signatures for places, helping AI models better understand the functional dynamics of the built environment.
GoogleAug 22Google Research introduced the Biomarker Discovery Framework, a multi-agent system that automates the prioritization of digital biomarkers from wearable sensor data. The framework uses an iterative loop of hypothesis generation, statistical analysis, and adversarial validation to identify candidates across mental health and metabolic domains. It identified 66 candidate biomarkers across 9,279 participant-observations while maintaining strict statistical rigor.
Google GemmaAug 22Google reports that its Gemma 4 31B model matches Claude Sonnet 5 on answer quality in AlphaSense benchmarks, while operating at roughly 40x lower cost. This cost-efficiency and low latency enable high-volume AI use cases that are uneconomical with larger frontier models. The benchmark evaluated 245 multi-step finance questions using pre-registered rubrics.
Google DeepMindAug 22Google DeepMind is partnering with Fenris Creations to use the EVE Online universe as a research testbed for frontier AI capabilities. The collaboration targets four open challenges: continual learning without forgetting, deep memory beyond current context windows, long-horizon planning over months, and complex multi-agent dynamics. This staged research program aims to discover breakthrough gameplay experiences and advance real-world AI.
AntigravityAug 22Google integrated Antigravity into Gemini Enterprise Standard and Plus subscriptions today. The update provides organization-wide access to agentic coding tools, including new IDE extensions for VS Code, Visual Studio 2026, JetBrains, and Zed. Administrators gain centralized governance through Workforce Identity Federation, audit logging, and granular budget controls, ensuring agentic workflows operate within secure enterprise boundaries.
AntigravityAug 22Google Antigravity launched IDE extensions for Visual Studio Code, Visual Studio, Zed, and JetBrains. The extensions integrate the agentic development platform into editors for code inspection, compiler diagnostic review, and localized adjustments. Enterprise teams access the tools via Gemini Enterprise or Google Cloud credentials, with data excluded from foundation model training and protected by organization-specific IAM policies.
GeminiAug 22Google provides eligible college students one year of premium Gemini plans at no cost, including Google AI Pro in the US and Google AI Plus in 140+ countries. The update adds global study tools to the Gemini app, featuring voice-driven Deep Research in Gemini Live, a student hub, study notebooks, and interactive visualizations. The offer runs through December 31, 2026.
Google AI StudioAug 22Google AI Studio Build now supports two-way synchronization with GitHub. The update adds support for importing existing repositories, pushing changes directly from AI Studio, and pulling updates from GitHub to maintain consistency across development environments. Commits now sync between the AI Studio build environment and external code repositories.
Google LabsAug 22Google Labs is expanding its experimental AI productivity agent, CC, to Australia and New Zealand. The agent now connects to Gmail to automatically create and maintain events in a dedicated Google Calendar. Additionally, invitations are rolling out to existing waitlist users in the US and Canada starting today.
GoogleAug 22Google AI Developers released a Chrome extension built in Antigravity with Gemini 3.7 Flash. The extension generates rich, visual definitions for highlighted text while browsing, powered by Nano Banana and Gemini Omni. This tool transforms web pages into interactive visual encyclopedias to support deeper learning.
GoogleAug 22Google Research introduced PhotoScan, a deep learning framework that estimates body composition metrics like body fat percentage and fat distribution from standard smartphone photos. The research demonstrates that these optical measurements predict insulin resistance with near-DXA accuracy, achieving an AUROC of 0.760. This prototype offers a scalable, non-invasive alternative to clinical scans for cardiometabolic risk assessment.
GoogleAug 22Google launched Sheets canvas, a Gemini 3.7 Flash-powered feature that transforms spreadsheets into live, interactive mini-apps. The tool generates functional UI components like sliders and dashboards, with changes syncing bi-directionally to the underlying sheet in real time. It is live for Google AI Pro and Ultra subscribers and rolling out to select Google Workspace business and education plans.
Google GemmaAug 22Google Developer Experts built a real-time AI Race Coach at Sonoma Raceway, running Gemma 4 locally on a Pixel 10. The system integrates Google Antigravity for telemetry orchestration and Gemini for cloud reasoning. This architecture delivers zero-latency audio coaching directly to drivers, processing data at 40 tokens per second to provide split-second track advice during high-speed racing.
GoogleAug 15Google announced the Pixel 11 series, Pixel Watch 5, and Pixel Tag, featuring the proactive Gemini Intelligence layer. The company also released Gemini 3.7 Flash, a model for coding and agents, at an introductory price of $0.75 per million input tokens. Additionally, Google DeepMind open-sourced WeatherNext 2, a tool for advanced cyclone track and intensity forecasting.
GeminiAug 14Google now allows the removal of visible watermarks from AI-generated images, videos, and music created with Nano Banana, Omni, and Lyria models. The new setting, available in Gemini and Flow, toggles visible overlays while maintaining mandatory invisible SynthID watermarks and C2PA metadata. This feature is rolling out over the coming days, with support for Search arriving soon.
AntigravityAug 13Google Antigravity has integrated Gemini 3.7 Flash, a model optimized for coding and multi-step agentic planning. The update delivers improved first-pass code accuracy and reasoning capabilities. Gemini 3.7 Flash is available at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026.
GeminiAug 13Google upgraded its Gemini Spark personal AI agent to run on the new Gemini 3.7 Flash model. This update improves the agent's precision and accuracy when executing multi-step workflows across Google Workspace apps. Gemini 3.7 Flash is available today for Google AI Pro and Ultra subscribers in over 160 countries.
Google DeepMindAug 13Google DeepMind launched Gemini 3.7 Flash, an agent-optimized model delivering improved performance in coding, knowledge work, and web development. The model achieves higher first-pass code accuracy and outperforms its predecessor on benchmarks like DeepSWE and WebDev Arena. It is available now in Google AI Studio and Android Studio at an introductory price of $0.75 per million input tokens.
AntigravityAug 12Google Antigravity launched Custom Agents, providing specialized, file-based configurations defined via Markdown and YAML frontmatter. These agents support execution symmetry as both main agents and subagents, scoped command-execution policies for autonomous tasks, and nested lifecycle hooks. Custom agents are available today in the Antigravity 2.0 desktop app and CLI, with IDE support arriving shortly.
GoogleAug 12Google Research introduced knowledge profiling, a framework that separately measures whether LLMs encode facts and whether they can recall them. Applying this to 13 models, researchers found that frontier LLMs encode nearly all facts but struggle with recall. Thinking-optimized models recover 40–65% of these inaccessible facts, suggesting that improving knowledge utilization is key to future factuality gains.
Google GemmaAug 12Google demonstrates Gemma 4 E2B running entirely locally on a Raspberry Pi 5 to power the Reachy Mini robot. The system handles voice, vision, and movement offline, achieving text generation speeds of 300 words per minute on 1.5 GB of memory. LiteRT splits workloads between the CPU and GPU to maintain real-time, low-latency performance.
GeminiAug 12Google is rolling out 14 new connected apps to Gemini over the next few weeks. Linking services like OpenTable, Ticketmaster, Wix, and Otter.ai in settings enables direct task management, trip planning, and project organization within the interface. The update adds integrations across productivity, local services, music, and home health categories.
Frequently asked questions