Google is the company behind Gemini, DeepMind research, Gemma open models, and AI across Search, Cloud, and Workspace. HeadsUpAI tracks Google across the AI ecosystem and curates every significant update — the latest being "Google Research Evaluates Conversational AI Agents for Symptom Assessment" (July 23, 2026) — so you get the whole story in a 30-second read.

What's new from Google?

The most recent Google update is "Google Research Evaluates Conversational AI Agents for Symptom Assessment" (July 23, 2026). HeadsUpAI curates every significant Google release as a 30-second read — what shipped and why it matters.

Google is the company behind Gemini, DeepMind research, Gemma open models, and AI across Search, Cloud, and Workspace. On this page you'll find every significant Google development HeadsUpAI has tracked recently — product updates, research, and company news — so you can keep up with where Google is heading without reading a dozen sources.

How often is Google news updated here?

Continuously. HeadsUpAI adds new Google updates as they're announced — usually within hours — and the 134 updates currently shown cover the past 90 days, newest first.

Google AI News & Updates — Latest Releases & Features

Q: What are the latest Google updates and releases?

The latest Google updates: "Google Research Evaluates Conversational AI Agents for Symptom Assessment", "Google Research Integrates Reinforcement Learning to Stabilize Quantum Computing", "Google DeepMind Commits $40M to DOE Genesis Mission for Science", "Google DeepMind Launches Gemini 3.5 Flash-Lite for High-Volume Tasks", and "Google Launches Gemini 3.5 Flash Cyber for Automated Vulnerability Remediation". HeadsUpAI has curated 134 Google updates over the last 90 days, covering product updates, research, and company news — listed newest first, presented straight, no hype, no bias.

GoogleJul 23

Google Research Evaluates Conversational AI Agents for Symptom Assessment

Google Research conducted a national-scale study of experimental conversational AI agents, built on Gemini Flash 2.0, for end-to-end symptom assessment. In a blinded evaluation, clinicians preferred the AI-generated differential diagnoses over human-provided ones in over 50% of cases. The study also identified correlations between AI-diagnosed respiratory infections and physiological shifts in wearable biosignal data.

GoogleJul 22

Google Research Integrates Reinforcement Learning to Stabilize Quantum Computing

Google Research integrated reinforcement learning with quantum error correction to enable continuous, autonomous calibration of quantum processors. By repurposing error-detection data as a learning signal, the system stabilizes the processor against hardware drift during computation. This approach improved logical stability 3.5-fold on the Willow processor and reduced logical error rates by 20% beyond conventional expert tuning.

Google DeepMindJul 22

Google DeepMind Commits $40M to DOE Genesis Mission for Science

Google DeepMind and Google Cloud are committing $40 million in AI tokens and cloud credits to the US Department of Energy’s Genesis Mission. This initiative provides researchers across all 17 DOE National Laboratories with access to frontier AI tools, including AlphaFold 3 and AlphaEvolve, to accelerate scientific discovery and autonomous experimentation across energy and security research.

Google DeepMindJul 22

Google DeepMind Launches Gemini 3.5 Flash-Lite for High-Volume Tasks

Google DeepMind released Gemini 3.5 Flash-Lite, a model optimized for high-throughput, repetitive tasks like data extraction. It delivers 350 output tokens per second and outperforms Gemini 3 Flash on agentic and coding benchmarks, including SWE-Bench Pro and OSWorld-Verified. The model is available now in the Gemini app, Google Search, and via API in Google AI Studio and Android Studio.

GoogleJul 21

Google Launches Gemini 3.5 Flash Cyber for Automated Vulnerability Remediation

Google launched Gemini 3.5 Flash Cyber, a specialized model built on 3.5 Flash for finding and fixing cybersecurity vulnerabilities. Integrated into the CodeMender agent, it achieves 83.1% on the CyberGym benchmark. The model will be available soon through a limited-access pilot program exclusively for government entities and trusted partners to address the speed of vulnerability discovery.

Google DeepMindJul 21

Google DeepMind Releases Gemini 3.6 Flash, 3.5 Flash-Lite, and Cyber

Google DeepMind launched three new Gemini models to scale agentic workflows. Gemini 3.6 Flash improves token efficiency by 17% at the same cost, while Gemini 3.5 Flash-Lite delivers 350 output tokens per second for high-throughput tasks. Gemini 3.5 Flash Cyber, a specialized model for vulnerability detection and patching, is now available through a limited-access pilot program in CodeMender.

Google AI StudioJul 21

Google Upgrades Gemini Batch API with Lower Latency and Partial Batches

Google upgraded its Gemini Batch API infrastructure, reducing p95 latency by 80% and p99 latency by 68%. The update improves reliability, achieving a batch success rate over 99.998% and a 98% reduction in batch expirations. Additionally, the API now supports partial batches, allowing for more flexible processing of large-scale requests.

Google GemmaJul 17

Google Gemma Releases gemma-trainer Skill for Agent-Led Model Fine-Tuning

Google Gemma released gemma-trainer, a new skill that automates local fine-tuning for Gemma 4 models. The skill enables AI agents to manage training configurations, execute supervised fine-tuning, and evaluate performance. It includes guardrails to prevent capability mismatches and supports exporting models for edge devices using LiteRT-LM.

GeminiJul 16

Google Gemini Adds Persistent Digital Avatars for Personalized Image Generation

Google Gemini launched a persistent digital avatar feature for creating custom images and videos without repeated selfie uploads. The avatar is invoked by typing a username in prompts after a one-time face and voice recording. This feature is available today for Google AI plan subscribers in select countries, excluding the EEA, Switzerland, and the United Kingdom.

Google GemmaJul 16

Google Gemma 4 31B Launches on LiveKit Inference for Voice Agents

Google Gemma makes the Gemma 4 31B model available on LiveKit Inference, optimized for real-time voice agents. The model delivers a 192ms time-to-first-token and 354ms time-to-first-audio. It achieves 76.9% accuracy on the tau2bench agentic tool-use evaluation, surpassing GPT-4.1. This integration provides a low-latency, cost-efficient option for production voice agent workloads.

Google DeepMindJul 16

Google DeepMind and Isomorphic Labs Detail New Proactive Bioresilience Framework

Google DeepMind and Isomorphic Labs published a joint approach to bioresilience, deploying frontier AI to prevent, detect, and respond to biological threats. The program includes threat modeling, AlphaEvolve-powered pathogen surveillance, and a dedicated unit for rapid medical countermeasure design. This effort leverages over 15 partnerships with government and biosecurity organizations to safeguard global health against future outbreaks.

GoogleJul 16

Google Research Links Diffusion Model Creativity to Neural Network Regularization

Google Research shows that diffusion model creativity is a mathematical consequence of neural network regularization. This process causes score smoothing, which forces the model to interpolate between training data points along the hidden data manifold. The findings, presented at ICLR 2026, explain how models generate novel data rather than simply memorizing training sets.

GeminiJul 15

Google Expands Gemini Spark Availability and Adds New Workspace Capabilities

Google is rolling out Gemini Spark to more countries and languages for Google AI Ultra subscribers. The update adds direct Google Doc editing, support for reading comments in Sheets and Slides, faster task execution, and parallel source retrieval for complex queries. These features increase the agent's performance and integration with Google Workspace.

Google GemmaJul 15

Google Gemma 4 Gains Flash Attention 4 and Vision Token Options

Google updated Gemma 4 with uniform Flash Attention 4 support on NVIDIA Hopper GPUs, increasing prefill throughput by 25-70% and reducing time-to-first-token by up to 31%. The release also adds manual vision token scaling up to 1120 for higher resolution, patches tool-calling consistency, and reduces model laziness to ensure more complete, accurate responses.

Google DeepMindJul 15

Google DeepMind Identifies Validation Bottleneck in AI-Driven Scientific Research

Google DeepMind published a policy essay identifying a growing validation bottleneck in scientific research. While AI agents rapidly generate hypotheses and experimental designs, physical testing remains slow and costly. The essay outlines four priorities for policymakers, including expanding access to agent tools, creating agent-ready data, investing in experimental infrastructure, and updating peer review processes to accommodate AI-generated research.

AntigravityJul 13

Google Antigravity Introduces Agent Teams for Parallel Autonomous Engineering Tasks

Google Antigravity launched Agent Teams, an orchestration layer that deploys specialized subagents to plan, build, and verify complex engineering tasks in parallel. Accessed via the /teamwork-preview command, the system uses Gemini 3.5 Flash to execute multi-step projects autonomously. It is available as a research preview for users on the Google AI Ultra $200/mo plan.

Google GemmaJul 10

Google Gemma Challenge Results: 5x Faster Inference via Agent Collaboration

Google Gemma and the community completed a six-day challenge where over 100 AI agents and humans collaborated to optimize Gemma 4 inference on NVIDIA A10G hardware. The effort achieved a 5x speedup, reaching 315 tokens per second in lossless mode. Agents demonstrated emergent coordination, including resource pooling and self-policing to prevent quality degradation during the optimization process.

Google AI StudioJul 10

Google AI Studio Adds Free Custom URLs for Deployed Apps

Google AI Studio now provides free custom URLs for deployed applications. Every app gains a personalized, memorable subdomain on the ai.studio domain during the publishing process. This update replaces auto-generated addresses with permanent, clean links for project sharing.

GeminiJul 10

Google Gemini Live Adds Real-Time Visualization and Google Maps Integration

Google updates Gemini Live with real-time project visualization powered by the Nano Banana image model. The feature generates images from camera input and integrates Google Maps for local search. These capabilities are available globally to all Gemini users at no cost.

Google GemmaJul 9

Google Gemma Fine-Tunes Gemma 4 to Translate Classical Korean Literature

Google Gemma demonstrates how to fine-tune the Gemma 4 E2B model on a single NVIDIA T4 GPU to translate Classical Korean into modern text. By using LoRA, the model’s character-by-character similarity score improved from 4.85% to 85.71%. This project provides a tutorial for preserving cultural history by making ancient texts accessible through lightweight model fine-tuning.

GoogleJul 9

Google Research Launches SensorFM Foundation Model for Wearable Health Data

Google Research introduced SensorFM, a foundation model trained on over one trillion minutes of unlabeled wearable data from five million participants. The model learns a general representation of human physiology that transfers across cardiovascular, metabolic, sleep, and mental health tasks. It also supports label-efficient adaptation and serves as a grounding tool for Personal Health Agents.

AntigravityJul 9

Google Antigravity CLI 1.1.0 Adds Interactive Execution and Review Workflows

Google Antigravity CLI 1.1.0 introduces interactive execution mode cycling, toggling between default, accept-edits, and plan modes via shift+tab. A new request-review default pauses before file writes, displaying line-level diff previews for approval. The update also consolidates mode commands, adds persistent settings in the /settings panel, and refines diff rendering and UI keybinding hints for agentic coding workflows.

Google GemmaJul 8

Google Brings Gemma 4 On-Device Support to React Native Apps

Google enables Gemma 4 to run fully offline in React Native applications using the react-native-executorch library. The integration supports local hardware acceleration via Vulkan on Android and MLX on Apple Silicon. This update allows mobile apps to perform on-device vision and tool-use tasks, such as reading documents and scheduling events, without cloud infrastructure.

Google AI StudioJul 8

Google AI Studio Adds One-Click GitHub Repository Import Feature

Google AI Studio launched an import from GitHub feature that automatically transforms repositories into a compatible format for its runtime. The integration supports immediate iteration, feature development, and deployment of existing codebases directly within the AI Studio Build environment.

GoogleJul 8

Google Research Expands FireSat Constellation with Three New Satellites

Google Research and the Earth Fire Alliance launched three new FireSat satellites from Vandenberg Space Force Base. This expansion scales the constellation’s ability to provide continuous, high-resolution wildfire detection. The mission aims to identify early-stage blazes as small as 5x5 meters, building on the architecture proven by the program's pilot satellite last year.

Google GemmaJul 7

Google DeepMind Releases Gemma 4 Technical Report Detailing New Architectures

Google DeepMind published the Gemma 4 technical report, detailing the architecture and efficiency of its latest open-weight multimodal models. The report covers the new encoder-free 12B variant, quantization-aware training, and multi-token prediction drafters. It also introduces a thinking mode that enables models to generate explicit reasoning traces before responding to prompts.

Google DeepMindJul 7

Google DeepMind Launches Predicting the Past Skill for Google Antigravity

Google DeepMind launched the Predicting the Past Skill for Google Antigravity, grounding Gemini in its expert Aeneas and Ithaca models. The skill converts complex epigraphic workflows like text restoration and chronological attribution into plain-English conversation. It supports large-scale analysis of ancient Greek and Latin inscriptions, demonstrated through case studies including Roman curse tablets and oracular records from Dodona.

GeminiJul 7

Google Gemini Spark Adds Real-Time Event Tracking and Automated Analysis

Google Gemini Spark now monitors topics and reacts to real-time events. The AI agent automatically generates and emails summaries based on specific triggers, such as match analyses after sports games. This feature is currently available in beta for Google AI Ultra subscribers in the United States, with an expansion to more countries planned later this month.

GeminiJul 6

Google Gemini Adds Google Business Profile Integration and Business Notebooks

Google Gemini now integrates with Google Business Profile to analyze performance data, draft review responses, and update profile information. New Business notebooks centralize business context, surfacing proactive alerts for unanswered customer questions and operational gaps. These features are available globally, excluding the EEA and the UK, to provide grounded, brand-specific assistance for business management.

Google DeepMindJul 6

Google DeepMind Partners With Apptronik to Train Gemini Robotics Models

Google DeepMind is partnering with Apptronik to use real-world data from the Apollo 2 humanoid robot to train Gemini Robotics. As Apptronik expands its 90,000-square-foot Robot Park facility in Austin, the data collected from bipedal and wheeled robot configurations will help refine foundational AI models for future humanoid systems.

GeminiJul 1

Google Expands Gemini Spark with macOS Integration and New Capabilities

Google brings Gemini Spark to the macOS app, enabling local task automation and remote execution via smart triggers. The update adds integrations with services like Canva and Dropbox, real-time topic tracking, and custom Model Context Protocol support. These features are rolling out today in beta for Google AI Ultra subscribers in the United States.

GoogleJun 30

Google Research Launches TabFM Foundation Model for Tabular Data

Google Research introduced TabFM, a foundation model for tabular classification and regression. The model uses a hybrid attention architecture trained on synthetic data to generate predictions on unseen tables in a single forward pass. It eliminates manual feature engineering and hyperparameter tuning, with direct integration into BigQuery ML via the AI.PREDICT SQL command coming in the next weeks.

NotebookLMJun 30

NotebookLM Expands Short Video Overviews to All Web Users

NotebookLM now offers Short Video Overviews to all users on the web in English. This feature generates 60-second vertical videos from uploaded sources, distilling complex information into quick visual explainers. The tool creates narrated slides that incorporate images, diagrams, and quotes directly from the user's documents.

Google DeepMindJun 30

Google DeepMind Launches Nano Banana 2 Lite and Gemini Omni Flash

Google DeepMind launched Nano Banana 2 Lite and Gemini Omni Flash. Nano Banana 2 Lite generates images in 4 seconds for $0.034 per 1K image, while Gemini Omni Flash supports video generation and conversational editing at $0.10 per second. Both models are available via Gemini API, AI Studio, and Enterprise Agent Platform, with chaining supported via the Interactions API.

NotebookLMJun 30

NotebookLM Launches 60-Second Short Video Overviews for Paid Subscribers

NotebookLM introduced Short Video Overviews, a new format that generates 60-second vertical videos from uploaded sources. This feature distills complex information into quick, visual explainers. It is currently available to Google AI Ultra and Pro subscribers on mobile and web, with support for English-language content. Free access is planned for the future.

Google AI StudioJun 28

Google AI Studio Adds Design Variations for One-Click UI Layout Generation

Google AI Studio launched Design Variations, a tool that generates and applies new UI layouts from natural language. This feature replaces manual aesthetic descriptions in prompts with one-click visual updates, maintaining existing content and structure during iteration. The tool is available now within the AI Studio interface.

GoogleJun 28

Google Research Accelerates Gemini Nano on Pixel with Frozen MTP

Google Research deployed a Multi-Token Prediction method to Pixel 9 and 10 devices that accelerates Gemini Nano v3 text generation by 50% or more. By attaching a lightweight prediction head to the frozen model, the architecture eliminates the memory overhead of separate drafter models while maintaining bit-exact output for features like AI Notification Summaries and Proofread.

AntigravityJun 28

Google Antigravity 2.0 Adds Guide Skill, Audio Support, and Search Improvements

Google Antigravity 2.0 version 2.2.1 introduces a built-in Antigravity Guide skill for contextual assistance. The update adds native audio file rendering for .mp3, .wav, .ogg, and .m4a formats within the sidebar and artifact viewer. Additionally, workspace file search now supports substring matching, removing the requirement for exact prefix queries to improve navigation reliability.

Google GemmaJun 28

Google Gemma 4 31B Launches on Cerebras with Virtual Hackathon

Google DeepMind released Gemma 4 31B on Cerebras, marking the platform's first multimodal model support with inference speeds reaching 1,500 tokens per second. To celebrate, Cerebras and Google DeepMind are hosting a 24-hour virtual hackathon starting June 28 with $5,000 in prizes. Participants receive early access to the model for the duration of the event.

GoogleJun 25

Google Research Introduces Linear Elastic Caching to Optimize Cloud Costs

Google Research introduced linear elastic caching, a cost-aware approach that dynamically adjusts cache size by framing page eviction as a ski rental problem. Using lightweight machine learning to predict optimal time-to-live values, the system reduced memory usage by 15.5% and total cost of ownership by approximately 5% in production tests on the Spanner database.

GeminiJun 25

Google Gemini Introduces Study Notebooks for Adaptive Personalized Learning

Google Gemini introduces study notebooks, an interactive space for adaptive learning. The feature accepts uploaded course materials and uses diagnostic quiz results to generate a plan of bite-sized lessons. A personalized dashboard tracks progress, with lessons updating automatically based on follow-up quiz performance. Study notebooks sync with NotebookLM and are available globally on the web at no cost.

GoogleJun 24

Google Research Identifies Mechanisms Behind Reasoning-Based Factual Recall in LLMs

Google Research published a study on how reasoning traces improve factual recall in LLMs, even for simple questions. The research identifies two mechanisms: a computational buffer effect and factual priming. It also demonstrates that filtering for hallucination-free reasoning paths during inference significantly increases model reliability and accuracy.

Google AI StudioJun 24

Google Integrates Computer Use as a Native Tool in Gemini 3.5 Flash

Google integrated computer use as a native tool into Gemini 3.5 Flash, replacing the previous standalone Gemini 2.5 computer use model. This update enables agents to interact with browser, mobile, and desktop environments while introducing optional enterprise safeguards for sensitive actions and prompt injection detection. Developers access the capability through the Gemini API and Gemini Enterprise Agent Platform.

Google GemmaJun 23

Google Gemma 4 26B A4B Achieves High Concurrency on DGX Spark

Google Gemma 4 26B A4B, quantized with NVFP4, sustains 16 parallel instances on a single NVIDIA DGX Spark. This configuration delivers 18 tokens per second per instance, with an aggregate throughput of 300 tokens per second. The architecture supports scaling up to 32 parallel runs, demonstrating efficient concurrent serving for multi-user and agentic workflows.

Google AI StudioJun 23

Google AI Studio Releases Comprehensive Getting-Started Guide for Interactions API

Google AI Studio released a getting-started guide for the Gemini Interactions API, a unified interface for models and agents. The guide covers setup, streaming, multimodal inputs, image and audio generation, structured output with Zod, Google Search grounding, function calling, and managed agents. It provides JavaScript examples for building autonomous workflows and long-running background tasks.

NotebookLMJun 23

NotebookLM Adds Full Customization to AI-Generated Study Flashcards

NotebookLM adds full customization to AI-generated flashcards. The update supports editing existing questions and answers, adding brand-new cards, and sharing final decks. This change shifts the tool from static AI generation to a flexible, curated study experience for refining educational content.

GoogleJun 22

Google Launches Interactions API for Gemini Models and Agents

Google released the Interactions API for Gemini models and agents into general availability. The unified interface features a stable schema, managed agent sandboxes provisioned by a single API call, and asynchronous background execution. It also adds native support for Computer Use, File Search, and Google Maps, alongside media generation capabilities including Nano Banana 2 and Lyria 3.

Google DeepMindJun 22

Google DeepMind and A24 Announce First-of-its-Kind AI Research Partnership

Google DeepMind and A24 are launching a multi-project research partnership to embed AI researchers directly into the studio's creative process. The collaboration aims to shape future AI tools through hands-on feedback from filmmakers. As part of the agreement, Google has also made an investment in the studio to support the development of new creative workflows and techniques.

Google GemmaJun 18

Google Gemma 3 Enables Autonomous Image Analysis on Orbiting Satellite

Google DeepMind’s Gemma 3 model is now running on Loft Orbital’s YAM-9 satellite, marking the first use of a vision-language model in orbit. The system, powered by NASA JPL software, autonomously identifies areas of interest from sensor data based on natural language queries. This capability allows satellites to triage data in space, reducing the need for ground-based human analysis.

Google DeepMindJun 18

Google DeepMind Details AI Control Roadmap for Securing Agentic Systems

Google DeepMind released its AI Control Roadmap, a defense-in-depth framework for managing advanced AI agents. The system treats agents as potentially misaligned, using supervisor AI to monitor reasoning and block harmful actions in real-time. DeepMind has already analyzed one million coding agent trajectories to refine these security protocols and currently employs live monitoring for its Gemini Spark agent.

Google GemmaJun 18

Google Gemma Demonstrates Local Multi-Agent Orchestration with Gemma 4 26B

Google Gemma published a cookbook demo showing Gemma 4 26B orchestrating 10 parallel sub-agents locally on macOS. The system uses an orchestrator to delegate tasks to concurrent instances, achieving speeds over 100 tokens per second. This architecture enables multi-agent workflows for SVG generation, code, and translation to run entirely on-device without cloud infrastructure.

Google DeepMindJun 17

Google DeepMind Partners With UK Government on AI Planning Prototype

Google DeepMind is co-developing an AI housing application planning prototype with the UK government to reduce processing times by up to 50%. The tool assists planning officers by consolidating data, identifying local policies, summarizing feedback, and drafting assessment reports. Early trials are underway in Barnet, Camden, and Dorset, with a national rollout planned for 2027.

Google GemmaJun 16

Google Gemma 4 E2B Now Runs on Intel AI PC NPUs

Google updated Gemma 4 E2B to support NPU acceleration on Intel Core Ultra processors via LiteRT and OpenVINO. This integration delivers 1.3x faster prefill performance and a 2.8x improvement in performance-per-watt compared to GPU execution. The update enables background LLM tasks on Intel AI PCs with zero thermal throttling or heavy battery drain.

AntigravityJun 15

Google Antigravity Now Generates Interactive 3D Modules From Napkin Sketches

Google Antigravity transforms hand-drawn napkin sketches into fully functional, interactive 3D learning modules. Powered by the Gemini 3.5 Flash model, the platform's agent autonomously researches the subject, creates an implementation plan, and writes the code for a browser-based application with interactive 3D scene exploration.

AntigravityJun 15

Google Antigravity Adds Android CLI and Skills via Plugin Bundles

Google Antigravity now offers curated Build with Google plugin bundles, including Android CLI and skills, to enable agentic Android development. These resources, configurable during onboarding or in Settings, allow agents to perform tasks like multi-device emulator testing and automated bug fixes. The platform also provides additional bundles for Firebase, Chrome DevTools, and scientific workflows.

GoogleJun 15

Google Research: AI Helps Identify Skin Conditions but Not Next Steps

Google Research published two studies showing that AI-assisted dermatology tools significantly improve a layperson's ability to name skin conditions. However, the research found no statistically significant improvement in a user's ability to determine appropriate medical next steps. These findings highlight that while AI aids identification, it does not automatically lead to better decision-making for health concerns.

GoogleJun 15

Google Research Introduces Gemini-SQL2 with 80.04% BIRD Benchmark Accuracy

Google Research introduced Gemini-SQL2, a text-to-SQL capability powered by Gemini 3.1 Pro. It achieved 80.04% execution accuracy on the BIRD benchmark, setting a new state-of-the-art for translating natural language into SQL queries that execute successfully. This capability addresses the challenge of generating accurate, runnable code from complex business contexts and natural language inputs.

Google LabsJun 15

Google Expands Project Genie Access to AI Ultra 5X Subscribers

Google Labs is expanding access to Project Genie, its experimental tool for generating interactive 3D worlds, to global Google AI Ultra 5X subscribers. The platform creates and renders navigable environments in real-time from text or image prompts. This update brings the world-building prototype to the 99.99 dollar monthly subscription tier introduced earlier this year.

GoogleJun 15

Google Research Debuts New Statistical Framework for Auditing Machine Unlearning

Google Research introduced Regularized f-Divergence Kernel Tests, a statistical framework for auditing machine unlearning and differential privacy. The method detects localized data shifts with higher sensitivity and fewer samples than traditional tools. It uses adaptive f-divergences to eliminate manual hyperparameter tuning, successfully identifying privacy violations in mechanisms that previously required millions of samples to detect.

GeminiJun 13

Google Expands Gemini Notebooks to EEA, UK, and Switzerland

Google expanded Notebooks in Gemini to the European Economic Area, United Kingdom, and Switzerland. Each notebook is a dedicated, focused space that keeps a project's sources, instructions, and chats in one place. Notebooks are now available to create in the Gemini app or on the web across these regions.

Google Research Evaluates Conversational AI Agents for Symptom Assessment

Google Research Integrates Reinforcement Learning to Stabilize Quantum Computing

Google DeepMind Commits $40M to DOE Genesis Mission for Science

Google DeepMind Launches Gemini 3.5 Flash-Lite for High-Volume Tasks

Google Launches Gemini 3.5 Flash Cyber for Automated Vulnerability Remediation

Google DeepMind Releases Gemini 3.6 Flash, 3.5 Flash-Lite, and Cyber

Google Upgrades Gemini Batch API with Lower Latency and Partial Batches

Google Gemma Releases gemma-trainer Skill for Agent-Led Model Fine-Tuning

Google Gemini Adds Persistent Digital Avatars for Personalized Image Generation

Google Gemma 4 31B Launches on LiveKit Inference for Voice Agents

Google DeepMind and Isomorphic Labs Detail New Proactive Bioresilience Framework

Google Research Links Diffusion Model Creativity to Neural Network Regularization

Google Expands Gemini Spark Availability and Adds New Workspace Capabilities

Google Gemma 4 Gains Flash Attention 4 and Vision Token Options

Google DeepMind Identifies Validation Bottleneck in AI-Driven Scientific Research

Google Antigravity Introduces Agent Teams for Parallel Autonomous Engineering Tasks

Google Gemma Challenge Results: 5x Faster Inference via Agent Collaboration

Google AI Studio Adds Free Custom URLs for Deployed Apps

Google Gemini Live Adds Real-Time Visualization and Google Maps Integration

Google Gemma Fine-Tunes Gemma 4 to Translate Classical Korean Literature

Google Research Launches SensorFM Foundation Model for Wearable Health Data

Google Antigravity CLI 1.1.0 Adds Interactive Execution and Review Workflows

Google Brings Gemma 4 On-Device Support to React Native Apps

Google AI Studio Adds One-Click GitHub Repository Import Feature

Google Research Expands FireSat Constellation with Three New Satellites

Google DeepMind Releases Gemma 4 Technical Report Detailing New Architectures

Google DeepMind Launches Predicting the Past Skill for Google Antigravity

Google Gemini Spark Adds Real-Time Event Tracking and Automated Analysis

Google Gemini Adds Google Business Profile Integration and Business Notebooks

Google DeepMind Partners With Apptronik to Train Gemini Robotics Models

Google Expands Gemini Spark with macOS Integration and New Capabilities

Google Research Launches TabFM Foundation Model for Tabular Data

NotebookLM Expands Short Video Overviews to All Web Users

Google DeepMind Launches Nano Banana 2 Lite and Gemini Omni Flash

NotebookLM Launches 60-Second Short Video Overviews for Paid Subscribers

Google AI Studio Adds Design Variations for One-Click UI Layout Generation

Google Research Accelerates Gemini Nano on Pixel with Frozen MTP

Google Antigravity 2.0 Adds Guide Skill, Audio Support, and Search Improvements

Google Gemma 4 31B Launches on Cerebras with Virtual Hackathon

Google Research Introduces Linear Elastic Caching to Optimize Cloud Costs

Google Gemini Introduces Study Notebooks for Adaptive Personalized Learning

Google Research Identifies Mechanisms Behind Reasoning-Based Factual Recall in LLMs

Google Integrates Computer Use as a Native Tool in Gemini 3.5 Flash

Google Gemma 4 26B A4B Achieves High Concurrency on DGX Spark

Google AI Studio Releases Comprehensive Getting-Started Guide for Interactions API

NotebookLM Adds Full Customization to AI-Generated Study Flashcards

Google Launches Interactions API for Gemini Models and Agents

Google DeepMind and A24 Announce First-of-its-Kind AI Research Partnership

Google Gemma 3 Enables Autonomous Image Analysis on Orbiting Satellite

Google DeepMind Details AI Control Roadmap for Securing Agentic Systems

Google Gemma Demonstrates Local Multi-Agent Orchestration with Gemma 4 26B

Google DeepMind Partners With UK Government on AI Planning Prototype

Google Gemma 4 E2B Now Runs on Intel AI PC NPUs

Google Antigravity Now Generates Interactive 3D Modules From Napkin Sketches

Google Antigravity Adds Android CLI and Skills via Plugin Bundles

Google Research: AI Helps Identify Skin Conditions but Not Next Steps

Google Research Introduces Gemini-SQL2 with 80.04% BIRD Benchmark Accuracy

Google Expands Project Genie Access to AI Ultra 5X Subscribers

Google Research Debuts New Statistical Framework for Auditing Machine Unlearning

Google Expands Gemini Notebooks to EEA, UK, and Switzerland

What is Google?

What's new from Google?

What are the latest Google updates and releases?

What does Google do?

How often is Google news updated here?