Gemini Embedding 2

Is this your product? Claim it free

Hidden from Google until claimed. Claimed listings appear in Google, carry your own description, and include 100 free custom questionnaires. The first 10 products also get a founding ranking slot with 100 review codes.

Start an assessment

TrustMark reading

Where this product sits

Gemini Embedding 2 is classified as Embedding models. Models that turn text or media into vectors for retrieval, not generative chat. Jobs: text embeddings; multilingual embeddings; multimodal embeddings.

Buyers typically come to Embedding models because retrieval quality is capped by a generic embedding the buyer cannot name. Classification is not a finding that this product solves that job.

This is TrustMark's classification reading for buyers comparing Embedding models products. It is not an Assurance Score, certificate, or ranked position.

What it is

Product classification

AI Model / API

Gemini Embedding 2 is the latest embedding model in Google's Gemini API and the first multimodal one. It turns text, images, video, audio and documents into vectors in a single shared embedding space, so content of one type can be searched, classified or clustered against another, across more than 100 languages. Google describes it as giving more accurate, context-aware results than keyword-based approaches, with common uses including semantic search, classification, clustering and retrieval-augmented generation, where embeddings help ground model answers. Developers can specify a task type to optimise the embeddings for the job. Unlike the older text-only gemini-embedding-001, which is still available, it produces one aggregated embedding when given several inputs, and the Batch API can be used to generate many at once. It is called through the Gemini API and Google's Gen AI SDKs.

Embedding models · Models & Foundation AI

What it does

How this product is used

Capabilities

Not specified

Sectors

Cross-industry / not specified

Deployment

Not specified

Autonomy

ASSISTIVE

Data exposure

Not specified

Category context

Why buyers look at Embedding models

Models that turn text or media into vectors for retrieval, not generative chat. Jobs: text embeddings; multilingual embeddings; multimodal embeddings.

Retrieval quality is capped by a generic embedding the buyer cannot name.

This is category context, not an independent assessment of Gemini Embedding 2. Vendor-claimed differences are listed above. Assessment results appear only after examination.

Browse Embedding models →

Where to get more information

• Product website · ai.google.dev/gemini-api/docs/embeddings
• Google website
• Company listing
• Ask to be invited to review
• Claim this listing to add a supplier statement

Other products from Google

• Agent Development Kit Claim

  Agent Development Kit (ADK) is an open toolkit for building production multi-agent systems, with support across Python, TypeScript, Go, Java, and Kotlin.

• Agent Platform Feature Store Claim

  Agent Platform Feature Store (formerly Vertex AI Feature Store) is Google Cloud's managed feature store for machine learning, part of Gemini Enterprise Agent Platform. It acts as a metadata and online serving layer over feature data kept in BigQuery, so teams do not copy data into a separate offline store. Features can be shared and reused through a feature registry, served for online predictions at low latency using Bigtable online serving, and tracked in Knowledge Catalog. It also supports embeddings and vector similarity search for nearest neighbours.

• Agent Platform Serverless Training Claim

  Serverless training in Gemini Enterprise Agent Platform (formerly Vertex AI Training) is Google Cloud's managed service for running large-scale machine learning training jobs. Training applications built on any ML framework run on Google Cloud infrastructure without provisioning or managing servers, and users pay only for the compute they consume. PyTorch, TensorFlow, scikit-learn and XGBoost have integrated support that simplifies preparing models for training and serving. Options include custom container settings, GPUs and Cloud TPU VMs, Spot VMs and reservations, persistent resources, distributed training, hyperparameter tuning and training pipelines.

• Agent Platform Vizier Claim

  Agent Platform Vizier (formerly Vertex AI Vizier) is Google Cloud's black-box optimisation service, part of Gemini Enterprise Agent Platform. It tunes the hyperparameters of machine learning models automatically, and can optimise any system with configurable parameters where evaluating a setting is expensive or there is no known objective function, for example finding the best network depth, width and learning rate for a TensorFlow model, or the most effective button colour and font size on a website. Users define a study configuration with the metrics to optimise and the parameters that affect them, and Vizier runs trials to find the best values.

• Agent Runtime Claim

  Agent Runtime (formerly Vertex AI Agent Engine) is the fully managed, serverless runtime in Google Cloud's Gemini Enterprise Agent Platform for deploying and scaling AI agents in production without managing infrastructure, with built-in observability. It works alongside Agent Platform Sessions and Memory Bank for conversation state and long-term memory, the Example Store and Evaluation Service for testing, monitoring and tracing agent behaviour, and sandboxed code execution. Agents can be built with Google's Agent Development Kit (ADK) or frameworks such as LangChain, LangGraph, AG2 and LlamaIndex, and provisioned with Terraform.

• Agent Search Claim

  Agent Search is Google Cloud's enterprise search and recommendations service, which Google is renaming from Vertex AI Search; earlier names include AI Applications, Vertex AI Search and Conversation, Enterprise Search and Generative AI App Builder. It lets developers, including those with limited machine learning skills, build Google-quality generative AI search and recommendation apps with natural language understanding, semantic search, AI-generated answers and summaries, and self-learning ranking models. Apps are set up in the Google Cloud console or through the Discovery Engine API and can be embedded on a website with a widget, with options optimised for media content and healthcare FHIR data.

• AI Commerce Search Claim

  AI Commerce Search is Google Cloud's product search, browse and recommendations service for retailers, formerly Vertex AI Search for commerce and now part of Gemini Enterprise for Customer Experience. Retailers import their product catalogue and shopper events, then serve search results, autocomplete, conversational product filtering and personalised recommendations on their own sites and apps. A Merchandising console lets teams boost, filter and control results. It is sold by Google LLC as a Google Cloud service.

• AppSheet Claim

  Google AppSheet is a no-code platform for building business apps and automations, including with Gemini assistance. Teams can create apps from prompts or data, capture forms and barcodes, automate notifications and workflows, and connect Google Workspace plus third-party systems under organisational governance controls.

• ARCore Claim

  ARCore is Google's augmented reality SDK, offering cross-platform APIs for building immersive experiences on Android, iOS, Unity and the web. Its fundamentals include motion tracking to position content relative to the world, anchors that keep track of an object's position over time, environmental understanding that detects the size and location of surfaces, depth understanding that measures distances between surfaces, and light estimation of the scene's average intensity and colour. The Geospatial API lets developers attach content remotely to areas mapped by Google Street View, and Google says ARCore supports world-scale experiences in over 100 countries.

• BigQuery ML Claim

  BigQuery ML is Google Cloud's capability for creating, running and managing machine learning and generative AI models inside BigQuery using SQL. Teams can train models on data in BigQuery, import models trained elsewhere in ONNX, TensorFlow, TensorFlow Lite or XGBoost format for in-database prediction, use data from multiple datasets, and integrate with Google Cloud's Agent Platform for wider AI and ML work.

• Chirp 3 Claim

  Chirp 3 is the latest generation of Google Cloud's multilingual automatic speech recognition models, available exclusively through the Cloud Speech-to-Text API V2 (model identifier chirp_3). Google says it improves accuracy and speed over earlier Chirp models and adds speaker diarization and automatic language detection. It supports streaming recognition for real-time audio, synchronous recognition for clips under a minute, and batch recognition for longer recordings. Transcription is generally available in the US and EU multi-regions for around 30 languages, including English (UK, US, India and Australia), French, German, Spanish, Hindi, Japanese and Korean, with many more languages in preview.

• Chirp 3: HD Claim

  Chirp 3: HD voices are the latest generation of voices in Google Cloud Text-to-Speech, powered by Google's newest generative models to deliver more realistic, emotionally expressive speech. Around 30 named voice options are available, such as Achernar, Charon, Kore, Puck and Zephyr, each with its own character. The voices cover a wide range of languages and locales, including English (UK, US, India and Australia), French, German, Hindi, Japanese, Arabic and many European and Indian languages, with Punjabi (India) and Chinese (Hong Kong) in preview. Developers can try them in Vertex AI Studio or call them through the Text-to-Speech API.

• Cloud Natural Language API Claim

  Cloud Natural Language API is a Google Cloud service that applies pre-trained machine learning models to text so developers can add natural language understanding to their applications. Features include sentiment analysis, entity analysis, entity sentiment analysis, content classification and syntax analysis. Typical uses include finding and labelling entities in emails, chats and social media posts, understanding customer opinion, and classifying content for recommendations and ad targeting. It can be combined with Speech-to-Text for audio, Vision API OCR for scanned documents and Translation API for multilingual text.

• Cloud Translation Claim

  Cloud Translation is Google Cloud's machine translation service. Translation API offers Basic and Advanced editions using Google's Neural Machine Translation, plus a Translation LLM and Adaptive Translation, document translation, and custom translation models, with translation into more than 130 languages. Translation Hub is an enterprise document translation platform; its Advanced tier adds translation memory, custom models, human review and machine translation quality prediction scores. Google states the first 500,000 characters sent to Translation API each month are free, with usage-based pricing after that.

• Cloud Video Intelligence API Claim

  Cloud Video Intelligence API is a Google Cloud service with pre-trained machine learning models that recognise over 20,000 objects, places and actions in stored and streaming video. It extracts metadata at the video, shot or frame level, which media teams use to index, search and filter video catalogues, and supports near real-time streaming annotation with event triggers based on detected objects. Google cites uses such as content discovery, highlight reels, recommendations and contextual advertising.

• Cloud Vision API Claim

  Cloud Vision API is Google Cloud's ready-to-use computer vision API, built on Google's pretrained machine learning models. Developers call it over REST or RPC to add common image analysis features to their applications, including image labelling, face and landmark detection, optical character recognition (OCR) and detection of explicit content. Each feature applied to an image is a billable unit, and the first 1,000 units each month are free. It is aimed at teams that want quick, simple integration of basic vision features without training their own models, and sits alongside Google Cloud's other vision offerings such as Document AI for document extraction and Video Intelligence API for video.

• Dialogflow CX Claim

  Dialogflow CX is Google Cloud's platform for building virtual agents that handle concurrent conversations with end users in text or audio. Its natural language understanding turns what users say into structured data that apps and services can use. Conversations are designed as flows, one per topic, each made of pages that act as states in a state machine, so large agents can be split across team members. Entity types (system or custom) control what data is extracted, parameters hold captured values, and page forms collect the details needed before moving on. Google notes some features are being renamed and generative playbook and flow features are moving to a single consolidated console.

• EmbeddingGemma Claim

  EmbeddingGemma is Google's family of lightweight, open-weight embedding models released under the Apache 2.0 licence. EmbeddingGemma 2, built on the Gemma 4 architecture, maps text, code, images, audio and video into a single 768-dimensional vector space, so developers can run multimodal retrieval-augmented generation, semantic search and zero-shot classification locally and offline on consumer GPUs or CPUs. It supports up to 8,192 tokens of text and code across more than 100 languages, and Matryoshka Representation Learning lets output be truncated to 512, 256 or 128 dimensions to cut storage. The earlier EmbeddingGemma 1 is a 308M-parameter multilingual text model designed for phones, laptops and tablets.

• Firebase Studio Claim

  Firebase Studio is Google's browser-based workspace for building full-stack apps with AI help. Developers can import existing repositories from GitHub, GitLab, Bitbucket or a local machine, start from framework templates, or use the App Prototyping agent to create a new app from natural language, mockups, drawings and screenshots. Gemini in Firebase helps with coding, debugging, testing, refactoring and documentation, and can take actions in the codebase. Workspaces can be shared, testers can preview apps through a URL, and built-in web previews and Android emulators support testing. Apps publish to Firebase App Hosting, Firebase Hosting, Cloud Run or custom infrastructure. During preview it offers 3 free workspaces, or up to 30 for Google Developer Program members.

• Flow Claim

  Google is a US technology company that sells consumer and cloud AI products, including Gemini, Search, and related developer APIs on Google Cloud. Flow is Google's software product, from Google LLC. Published at labs.google/fx/tools/flow. Directory filing: Storyboarding. Vendor legal name on this product record: Google LLC. Flow is Google's software product, from Google LLC. Published at labs.google/fx/tools/flow. Directory filing: Storyboarding.

• Gemini Claim

  Gemini is Google's consumer and Workspace conversational assistant for writing, planning, image understanding and related tasks, offered on the web and in Google apps. Google markets free access plus paid Google AI plans. It is a hosted Google assistant, not an on-premise model package. Gemini is Google's software product, from Google LLC. Published at gemini.google.com. Directory filing: Multimodal assistants. Deployment options: SaaS and Mobile / on-device. Recorded capabilities: Generation, Vision, and Voice generation.

• Gemini 2.5 Flash Claim

  Gemini 2.5 Flash is a Flash-tier model in Google's Gemini API that Google describes as its best model for price-performance, with well-rounded capabilities. It is aimed at large-scale processing, low-latency and high-volume tasks that need thinking, and agentic use cases. It accepts text, image, video and audio input and returns text, with an input limit of about one million tokens and up to 65,536 output tokens. It supports thinking, function calling, structured outputs, code execution, caching, file search, URL context, and grounding with Google Search and Google Maps, and can run through batch, flex and priority inference. Google says the 2.5 models are not deprecated and will keep being served until further notice, but access is now limited to users who have used them before, and it recommends newer models such as 3.8 Flash for new projects.

• Gemini 2.5 Pro Claim

  Gemini 2.5 Pro is a thinking model in Google's Gemini API, which Google describes as able to reason over complex problems in code, maths and STEM, and to analyse large datasets, codebases and documents using long context. It accepts audio, image, video, text and PDF input and returns text, with an input limit of about one million tokens and up to 65,536 output tokens. It supports thinking, function calling, structured outputs, code execution, caching, file search, URL context, and grounding with Google Search and Google Maps, and can run through batch, flex and priority inference. Google says the 2.5 models are not deprecated and will keep being served until further notice, but access is now limited to users who have used them before, and it recommends its latest models for new projects.

• Gemini 3.6 Flash Claim

  Gemini 3.6 Flash is a Flash-tier model in Google's Gemini API, built to give strong reasoning at higher speed and lower cost for real-world work. Google positions it for the agentic era, saying it does well at code generation, agentic execution and spatial reasoning, and is particularly effective for fast agent loops that iterate through complex coding cycles. It accepts text, image, video, audio and PDF input and returns text, with an input limit of about one million tokens and up to 65,536 output tokens. It supports function calling, structured outputs, thinking, code execution, caching, file search, URL context, and grounding with Google Search and Google Maps, with computer use in preview, and can run through batch, flex and priority inference. Google now lists Gemini 3.8 Flash as its newest Flash model, while 3.6 Flash remains available for production use, and requests to the deprecated Gemini 3.5 Flash are routed to it.

• Gemini Cloud Assist Claim

  Gemini Cloud Assist is Google Cloud's AI assistant for cloud operations, powered by Gemini 3. It helps teams across the application lifecycle, from design and deployment to monitoring, troubleshooting and cost optimisation. A multi-agent system works through complex operational goals and only acts with explicit user authorisation. It summarises logs and explains errors, turns natural-language intent into architecture diagrams and Terraform, gcloud or kubectl blueprints through Application Design Center, gives security and compliance guidance such as IAM role recommendations, and links with Cloud Hub to act on application events and cost recommendations. It works in the Google Cloud console, Gemini CLI and IDEs. Google says it is currently in preview and free of charge, with some features to be charged at general availability.

• Gemini Code Assist Claim

  Gemini Code Assist Standard and Enterprise are Google's AI coding assistants for teams and businesses, built on Gemini 3 with a 1 million token context window. They offer code completion and generation in IDEs such as Visual Studio Code and JetBrains IDEs, natural language chat, an agent mode in preview for multi-file edits with human oversight and MCP tool integration, and access to the open-source Gemini CLI in the terminal. Code customisation can tailor suggestions to an organisation's private codebases, and admins get usage metrics and dashboards. For the free individual tier and Google One users, Google replaced the Gemini CLI and Code Assist IDE extensions with Antigravity on 18 June 2026.

• Gemini Enterprise Claim

  Gemini Enterprise is Google Cloud's work agent for businesses, which Google now markets as Gemini for Business. It is a single agent that works across a company's email, documents, code, data and business apps, remembering tasks, facts and methods across devices so staff can hand off work rather than supervise it. Long-running jobs run on their own secure machine in Google Cloud. It runs under each person's existing permissions, keeps a full audit trail, and lets admins set budget caps per project. It can route jobs to Google, third-party, open or custom-trained models under one set of permissions and one bill. Google offers role-specific skills and connectors for sales, marketing, engineering, financial services and legal work, and a 30-day no-cost trial of the Gemini Enterprise Plus edition.

• Gemini Enterprise Agent Platform Claim

  Gemini Enterprise Agent Platform (formerly Vertex AI, including Vertex AI Agent Builder) is Google Cloud's platform for building, scaling, governing and optimising enterprise AI agents and applications grounded in an organisation's own data. It offers more than 200 Google, third-party and open models through Model Garden, including Gemini, Anthropic's Claude family and Gemma, with tuning options and a Model Evaluation service for assessing generative AI models. Data science teams can also train, tune and deploy machine learning models on the same platform, and Google Antigravity is available through it to steer and orchestrate multiple agents across workflows.

• Gemini Enterprise Agent Platform Vision Claim

  Google is a US technology company that sells consumer and cloud AI products, including Gemini, Search, and related developer APIs on Google Cloud. Gemini Enterprise Agent Platform Vision is Google's platform, from Google LLC. Published at docs.cloud.google.com/vision-ai/docs/overview. Directory filing: Visual inspection. Gemini Enterprise Agent Platform Vision is Google's platform, from Google LLC. Published at docs.cloud.google.com/vision-ai/docs/overview. Directory filing: Visual inspection.

• Gemini for Google Workspace Claim

  Gemini for Google Workspace is the set of AI features Google builds into its business apps. Google describes Gemini in Gmail summarising threads, finding key details and drafting emails; in Meet taking notes so people can focus on the conversation; in Docs creating formatted first drafts that match a user's tone; in Sheets building spreadsheets and filling tables from plain-language prompts; in Slides adding editable slides to decks; and in Drive finding and summarising information across files. Business plans also include the Gemini app and Gemini Notebook for research grounded in a user's sources, Studio for building automated flows, and Vids for making videos. Workspace Intelligence brings personal context into these apps. Google says company data stays the customer's, with admin controls over access, and is not used to train models.

• Gemini Nano Claim

  Gemini Nano is Google's on-device foundation model for Android, aimed at generative AI features that need to work without a network connection or without sending data to the cloud, where low cost and privacy matter most. It runs in Android's AICore system service, which uses the device hardware for low-latency inference, manages model downloads and updates, and applies safety filters. Developers reach it through ML Kit's GenAI APIs, which cover prompting with text or images, summarisation, proofreading, rewriting, image description and speech recognition. Google says AICore follows Private Compute Core principles, has no direct internet access and keeps no record of inputs or outputs. For devices without Gemini Nano, the Firebase AI Logic hybrid API can fall back to Gemini models in the cloud.

• Gemini Robotics 2 Claim

  Gemini Robotics 2 is Google DeepMind's most advanced vision-language-action (VLA) model, which turns what a robot sees and the instructions it is given into motor control. Google says it can carry out a variety of tasks it was not specifically trained on and adapt to unfamiliar situations as they arise. It can be adapted to a bi-arm robot in a few hours and scales up to full humanoids, with whole-body control from feet to fingertips and the dexterity for delicate actions such as screwing in a light bulb or tying knots. It also coordinates several robots working in a shared space, and works alongside Gemini Robotics ER 2, which handles reasoning and planning, while a lighter On-Device 2 version runs locally on robot hardware.

• Gemini Robotics ER 2 Claim

  Gemini Robotics ER 2 is Google DeepMind's embodied reasoning model, described as a high-level brain for robots. It understands the physical world, identifies objects and movement precisely, plans multi-step tasks and then hands motor execution to a lower-level vision-language-action model, thinking about the next step while the robot is still acting. It can coordinate several robots on a shared task, tracks whether a task succeeded or needs another attempt, and can call tools such as Google Search or Google Calendar. It follows everyday spoken commands and can explain its reasoning. It accepts text, image, video and audio input and returns text, and is in public preview through Google AI Studio and the Gemini API, with a private preview on Gemini Enterprise Agent Platform.

• Gemma 3n E4B Claim

  Gemma 3n E4B is the larger of Google's two Gemma 3n open-weight models, designed to run on everyday devices such as phones, laptops and tablets. The E stands for effective parameters: through parameter skipping and Per-Layer Embedding caching to fast local storage, it runs with a much smaller memory load than its total parameter count, and its MatFormer architecture lets developers activate only the parameters a request needs. It handles text, images and audio input, with a MobileNet-V5 vision encoder and audio support for speech recognition and translation, and vision and audio parameters can be skipped when not needed to save memory. It was trained on more than 140 languages and offers a 32K token context. The weights are licensed for responsible commercial use and are available on Kaggle and Hugging Face.

• Gemma 4 E2B Claim

  Gemma 4 E2B is the smallest model in Google's Gemma 4 family of open-weight models, with 2 billion effective parameters and built for mobile, edge and in-browser use on devices such as Pixel phones and Chrome. Like the rest of the family it is designed as a reasoning model with configurable thinking modes, built-in function calling and native system prompt support. It handles text and images with variable resolution, plus video, and is one of the Gemma 4 sizes with native audio input. It has a 128K context window and includes a draft model for faster speculative decoding. Google lists about 1.1 GB of memory for its mobile build. The weights permit responsible commercial use and can be downloaded from Kaggle and Hugging Face.

• Gemma 4 E4B Claim

  Gemma 4 E4B is a small model in Google's Gemma 4 family of open-weight models, with 4 billion effective parameters and aimed at mobile, edge and in-browser deployment where a little more capability than E2B is needed. It is built as a reasoning model with configurable thinking modes, built-in function calling for agent use and native system prompt support. It processes text and images with variable resolution, plus video, and has native audio input. It offers a 128K context window and a draft model for faster speculative decoding, and Google lists about 2.5 GB of memory for its mobile build. The weights permit responsible commercial use, so teams can tune and deploy it in their own products, and are available on Kaggle and Hugging Face.

• Google Assistant Claim

  Google Assistant is Google's voice assistant, started by saying "Hey Google". Google offers it on phones, tablets, smart speakers, smart displays, TVs, watches and in cars. It can send texts, play music and playlists, and control smart home devices such as lights and thermostats, and Google pairs it with the Google Home app for setting up a connected home. On Android phones it works with partner apps such as Spotify, Discord, Etsy, Nike Run Club, MyFitnessPal and Walmart, so tasks can be started in those apps by voice. Google says Assistant has built-in privacy controls that let users choose how their information is used. Some content may need a subscription.

• Google Cloud Agent Assist Claim

  Agent Assist is Google Cloud's AI tool for customer service teams, now part of Gemini Enterprise for Customer Experience. It gives contact centre representatives in-the-moment coaching and next-best-action guidance, suggests ready-to-send responses, and answers customer questions from a central knowledge base drawn from websites, documentation, CRM and other sources. It transcribes calls in real time and uses Gemini models to write post-interaction summaries with key takeaways and next steps, cutting after-call work. An AI Trainer feature runs interactive training simulations for new hires. Google says Agent Assist helps representatives handle 28% more conversations and respond 15% faster with Smart Reply.

• Google Cloud Confidential Computing Claim

  Google Cloud Confidential Computing encrypts data while it is in use, so it stays private even while being processed in the cloud. Offerings include Confidential VMs, including C3 machines with Intel AMX acceleration for AI and A3 machines with NVIDIA H100 GPUs for protected AI and machine learning workloads; Confidential GKE Nodes, which keep Kubernetes data encrypted in memory with processor-generated keys unavailable to Google; Confidential Space, a trusted execution environment for joint data analysis and model training without exposing data to other parties or operators; and Confidential Dataflow and Dataproc for data pipelines with inline memory encryption.

• Google Cloud Document AI Claim

  Google Cloud Document AI is a document processing platform for extracting, classifying and splitting data from documents. Its generative AI powered custom extractor pulls structured data out of documents out of the box and can be fine-tuned with as few as 10 documents; the custom splitter separates composite files such as mortgage packages into single-class documents, and the custom classifier sorts documents into your own classes. Enterprise Document OCR extracts text and layout. Processors are created in the console UI or by API and connect to BigQuery and other Google Cloud services.

• Google Coral Claim

  Coral is Google's full-stack platform for edge AI, combining an AI-first, standards-based hardware architecture with a unified developer toolchain so AI can run locally on devices. Software developers can bring PyTorch, JAX or LiteRT models to edge devices using MLIR-based compiler toolchains and simulators. Hardware developers can use reference designs and Coral's open-source, RISC-V based architecture to add high-performance, ultra-low-power AI to their own silicon. Google lists a partner ecosystem to help bring Coral-based products to market.

• Google Document AI Form Parser Claim

  Form Parser is a pretrained processor in Google Cloud's Document AI that pulls structured data out of forms and other documents. It extracts key-value pairs (such as a label and the value written next to it), simple tables, checkboxes with whether they are ticked, plain text and layout, and 11 generic fields out of the box, including names, addresses, dates, email addresses, phone numbers, prices and IDs. It uses Google's latest OCR engine on images and scanned PDFs, supports over 200 languages, and runs in eight regions. It cannot be trained on your own documents, so it suits teams handling high volumes of conventional forms who want extraction without building and maintaining their own parsers; Google points to its custom extractor where trained extraction is needed.

• Google Earth Engine Claim

  Google Earth Engine is a cloud-based platform for planetary-scale Earth science data and geospatial analysis. It combines a public catalogue of more than eighty petabytes of satellite imagery and geospatial datasets, covering over thirty years and updated daily, with analysis that runs on Google's cloud. Users work through Python and JavaScript APIs or a web-based code editor to detect change, map trends and quantify differences on the Earth's surface, and can use the Satellite Embedding Dataset produced by the AlphaEarth Foundations model. Earth Engine is available for commercial use and is free for academic and research use.

• Google Gen AI SDK Claim

  The Google Gen AI SDK is Google's open-source client library for building with its generative models. The Python version, published on GitHub as googleapis/python-genai, gives developers one interface to both the Gemini Developer API and Google's enterprise agent platform on Google Cloud, formerly Vertex AI, and Google's Gemini documentation uses it alongside JavaScript, Java and Go versions in its code samples. It covers content generation, chat, embeddings, the Live API for real-time sessions, video generation and automatic function calling. Google notes that the next major version changes how automatic function calling works, moving it to the chat modules, and recommends pinning below version 3.0.0 to avoid unexpected changes. It also points developers using AI coding assistants to its Gemini API skills repositories so generated code follows current SDK patterns.

• Google Nest Cam (indoor, wired) Claim

  Nest Cam Indoor (wired, 3rd gen) is Google's plug-in indoor security camera. It records 2K video with HDR, and Google says it keeps colours vivid in low light and has its widest, tallest field of view yet, with zoom and crop. Built-in intelligent alerts flag people, animals and vehicles, Activity Zones limit alerts to the areas that matter, and six hours of 10-second event previews are included. With a Google Home Premium subscription, sold separately, Gemini adds searchable video history, detailed event descriptions, a daily recap of recorded events, and up to 10 days of 24/7 video history. The feed can be viewed in the Google Home app and on compatible devices such as Google TV Streamer, Nest displays and Pixel Tablet. Google cites encrypted video, two-step verification and a green LED that shows when the camera is streaming.

• Google Pixel 11 Pro Claim

  Google Pixel 11 Pro is Google's flagship phone, sold in 6.3-inch Pro and 6.8-inch Pro XL sizes, and Google says it is designed for Gemini Intelligence, which gives proactive, personal help with everyday tasks across apps. It runs on the Google Tensor G6 chip with a vapor chamber to keep speeds up, and the Titan M3 security chip helps protect data. Two upgraded cameras include a larger telephoto sensor with 30% better light sensitivity for modes such as Instant Night Sight, and Magic Capture records a video and photos with one tap, then edits them into a collection ready to share. HiLight glows when the phone is face down and a favourite contact gets in touch or Gemini is talking. Google quotes 30+ hours of battery, wireless charging up to 29% faster, and built-in Pixelsnap that works with MagSafe accessories.

• Google Pixel Buds Pro 2 Claim

  Pixel Buds Pro 2 are Google's premium wireless earbuds and the first to run on Google's own Tensor A1 chip, which Google says powers twice the noise cancellation of earlier Pixel Buds. Adaptive Audio turns down loud surroundings while keeping you aware of them, Active Noise Cancellation with Silent Seal 2.0 blocks outside sound, and Loud Noise Protection dampens sudden bursts such as sirens. They offer 11 mm drivers, spatial audio with head tracking, a twist-to-adjust stabiliser for workouts, and up to 30 hours of listening with the charging case. Gemini Live can be used hands-free to brainstorm, add notes or schedule events, and the Pixel Buds app adds a five-band EQ. Google aims them at Android and Pixel users who want all-day earbuds for work, exercise and travel.

• Google Pixel Watch 5 Claim

  Google Pixel Watch 5 is Google's latest smartwatch, offered in 41 mm and 45 mm sizes and as a Stephen Curry special edition. Google calls it its most comprehensive health and wellness tracker, with blood pressure and insulin resistance trends, loss of pulse detection that can prompt a call to emergency services, an ECG app for checking signs of AFib, and sleep tracking that produces a personalised Sleep Score. For fitness it has dual-frequency GPS and heart rate tracking across more than 50 activities. Gemini can be used by raising your wrist, and Google says the watch is 20% faster. It has an Actua 360 domed display with 3,000 nits peak brightness, up to 40 hours of battery, IP68 water and dust resistance, and an LTE option. Personalised coaching comes with a Google Health Premium subscription.

• Google Search Claim

  Google is a US technology company that sells consumer and cloud AI products, including Gemini, Search, and related developer APIs on Google Cloud. Google Search is Google's software product, from Google LLC. Published at google.com. Directory filing: AI search engines. Vendor legal name on this product record: Google LLC. Google Search is Google's software product, from Google LLC. Published at google.com. Directory filing: AI search engines.

• Google Security Operations Claim

  Google Security Operations, formerly known as Chronicle, is Google Cloud's security operations platform. It brings SIEM, SOAR and threat intelligence into one experience so security teams can collect security telemetry, apply threat intelligence to identify high-priority threats, and drive response through playbook automation, case management and collaboration. Google positions it for SIEM migration and security operations centre modernisation, and for national-scale government cyber defence through Google Cloud Cybershield. It sits within the Google Unified Security portfolio alongside recommended partner integrations.

• Imagen Claim

  Imagen is Google's family of image generation models, offered to developers on Vertex AI, Google Cloud's AI platform, alongside Gemini's own image models. With it, teams can generate new images from a text prompt, edit or extend an uploaded or generated image within a masked area they define, and upscale existing, generated or edited images. It can be tried in Vertex AI Studio or called through the API, with Imagen 3 and Imagen 4 among the model versions listed. Google aims it at application developers building products that turn written descriptions into finished visual assets in seconds.

• Instant Custom Voice Claim

  Instant Custom Voice is a Chirp 3 feature in Google Cloud Text-to-Speech that creates a personalised voice model from high-quality audio recordings of a speaker. Once created, the custom voice can be used to synthesise speech through the Cloud Text-to-Speech API, which supports streaming and long-form text. Google says it supports text-based prompting, where punctuation, pauses and disfluencies add natural pacing, pace control from 0.25x to 2x speed, and experimental pause tags and custom pronunciations using IPA or X-SAMPA. Voices cloned in US English can also produce speech in several other locales, including German, Spanish, French and Brazilian Portuguese. Speakers must record a consent statement in their language confirming they own the voice, and the feature is offered in many languages including UK English, with global, US, EU, London and Asian regional endpoints. Access is restricted to allow-listed customers, who request it through Google's sales team.

• JAX Claim

  JAX is an open-source Python library from Google for accelerator-oriented array computation and program transformation, built for high-performance numerical computing and large-scale machine learning. It offers a NumPy-style API and composable transformations for just-in-time compilation (jax.jit), automatic differentiation (jax.grad), batching (jax.vmap) and parallelisation, including sharding across many devices. The same code runs on CPU, GPU and TPU, and Pallas lets developers write custom GPU and TPU kernels. For neural network training, the project points users to the JAX AI Stack.

• Jules Claim

  Jules is Google's autonomous coding agent. A developer selects a GitHub repository and branch and writes a prompt, or adds the "jules" label to a GitHub issue; Jules clones the repository to a cloud VM, drafts a plan using Gemini models, shows a diff of its changes for approval, and creates a pull request. Typical tasks include bug fixing, version bumps, writing tests and building features. The free plan allows 15 tasks per day and 3 concurrent tasks; Jules in Pro raises this to 100 tasks per day and 15 concurrent, and Jules in Ultra to 300 tasks per day and 60 concurrent, with higher or priority access to the latest Gemini models.

• Keras Claim

  Keras is an open-source deep learning API from Google, designed to keep model code small, readable and easy to debug and deploy. Keras 3 is multi-backend, so the same model code runs on JAX, TensorFlow or PyTorch and can move between those frameworks. Developers build models with the Sequential or Functional APIs and train them with a few lines of code, and the companion KerasHub library provides pretrained presets such as Gemma language models and Stable Diffusion image generation.

• Learning Interpretability Tool Claim

  The Learning Interpretability Tool (LIT) is an open-source, visual and interactive model-understanding tool from Google Research's PAIR team for researchers and practitioners. It helps answer questions such as which examples a model performs poorly on, why it made a prediction, and whether it behaves consistently when text style, tense or pronouns change. Built-in capabilities include salience maps, metrics, counterfactual generation, model and datapoint comparison, embedding visualisation and TCAV. LIT supports classification, regression, text generation and other tasks on text, with preliminary image and tabular support, works with TensorFlow and PyTorch, and runs as a standalone server or in Colab, Jupyter or Vertex AI notebooks.

• MedGemma 27B Multimodal Claim

  MedGemma 27B Multimodal is the largest model in Google's MedGemma collection, open models for medical text and image comprehension built on Gemma 3 and published through Health AI Developer Foundations. It is aimed at developers building healthcare applications, with common uses including medical image interpretation such as drafting image reports or answering questions about images, medical text comprehension, and clinical reasoning tasks like patient interviewing, triage, decision support and summarisation. Google says the 27B models generally give the best performance for text-based medical tasks. Google is clear that MedGemma is not yet clinical-grade: developers must validate it for their intended use and will usually need to adapt it with prompting, fine-tuning or agentic orchestration before production.

• Model Armor Claim

  Model Armor is a Google Cloud security service that screens AI prompts, responses and agent interactions at runtime. It protects against prompt injection and jailbreak attempts, helps prevent leakage of personal, financial, credential and custom sensitive data in prompts and model responses, and detects malicious files, malware and unsafe URLs, including indirect prompt injection hidden in otherwise legitimate content. Google says it combines rules-based controls, machine learning models and AI reasoning models. It is model and cloud agnostic through a REST API, covering Gemini, OpenAI, Anthropic, Llama and other models, and also offers no-code in-line protection integrated with Google Cloud services.

• Nest Learning Thermostat (4th gen) Claim

  The Nest Learning Thermostat (4th gen) is Google's most advanced smart thermostat. Its Smart Schedule learns the temperatures a household prefers and when, then suggests small changes in the Google Home app to help save energy. It comes with a Nest Temperature Sensor to balance hot and cold rooms, and features such as Adaptive Eco and natural heating and cooling use occupancy and outdoor conditions to cut wasted energy, with Google citing savings of up to 31% on heating and cooling bills. A large borderless display shows the time, temperature or weather from across the room, and System Health Monitor can flag problems with the heating and cooling system early. Google says it works with most HVAC systems and is designed for self-installation through the Google Home app.

• NotebookLM Claim

  NotebookLM is Google's source-grounded research notebook assistant for synthesising and chatting with user documents. Public marketing also emphasises source-grounded answers from user docs. It is presented for Knowledge workers and research teams. NotebookLM is Google's software product, from Google LLC. Published at notebooklm.google. Directory filing: Research assistants.

• Perspective API Claim

  Perspective API is a machine learning API from Jigsaw, a unit within Google, that scores comments for attributes such as toxicity to help publishers and platforms moderate online conversations. Jigsaw has announced that Perspective API is being sunset: the service remains active until 31 December 2026 and will not be in service after 2026, requests for usage and increased quota closed in February 2026, and no direct migration support is offered.

• Project Relate Claim

  Project Relate is an Android beta app from Google Research that helps people with non-standard speech be understood. Each user records 500 phrases so the app can learn their speech patterns, and after a few days of processing it creates a personalised speech recognition model. Its Listen feature transcribes the user's speech as live captions, Repeat restates what they said in a clear computerised voice for short face-to-face conversations, and the Relate keyboard brings speech-to-text into any app. Custom Cards let users teach it important names, places and phrases. Google says the app is not accepting new users at the moment, although existing users can still use their models.

• Route Optimization API Claim

  Route Optimization API is a Google Maps Platform API that builds optimised route plans for one or many vehicles and their stops, using Google Maps data and the business's own parameters. Teams set objectives such as travel efficiency, on-time arrival, lower vehicle use cost and fair load balancing across the fleet, alongside constraints such as driver working hours and breaks, vehicle capacity, delivery time windows and task dependencies. It offers synchronous and asynchronous requests and returns detailed plans with assigned tasks, vehicle routes and performance metrics. Google aims it at logistics companies planning next-day deliveries, on-demand delivery services choosing which driver should take a new order, and field service providers sequencing appointments and re-planning routes when things change.

• SigLIP 2 So400m patch14 384 Claim

  SigLIP 2 So400m patch14 384 is an open-weight vision-language encoder from Google, published on Hugging Face under the Apache 2.0 licence. SigLIP 2 builds on the original SigLIP sigmoid-loss pretraining with extra training objectives, including a decoder loss, global-local and masked prediction losses, and adaptability to different aspect ratios and resolutions, which Google says improves semantic understanding, localisation and dense features. This checkpoint uses the shape-optimised So400m architecture at 384-pixel resolution with 14-pixel patches, about 1.1 billion parameters in total. It can be used directly for zero-shot image classification and image-text retrieval, or as the vision encoder inside a larger vision-language model. It was pre-trained on Google's WebLI dataset using up to 2,048 TPU v5e chips, and runs with the Hugging Face Transformers library.

• SynthID Claim

  SynthID is Google DeepMind's watermarking tool for AI-generated content. It embeds imperceptible digital watermarks directly into AI-generated images, video, audio and text across Google's generative AI consumer products. Image and video watermarks are added at the moment of creation and are designed to survive cropping, filters, frame-rate changes and lossy compression; audio from the Lyria music model and NotebookLM podcast feature is watermarked inaudibly; and text from the Gemini app and web experience is watermarked by adjusting token probability scores without affecting output quality. Users can upload an image, video or audio clip to Gemini, Search or Chrome and ask whether it was created or altered by Google AI.

• TensorBoard Claim

  TensorBoard is TensorFlow's open-source visualisation toolkit for machine learning experiments. It tracks and plots metrics such as loss and accuracy, visualises the model graph of ops and layers, shows histograms of weights, biases and other tensors as they change during training, projects embeddings into lower-dimensional space, displays image, text and audio data, and profiles TensorFlow programs to help find performance bottlenecks.

• TensorFlow Claim

  TensorFlow is Google's open-source end-to-end machine learning platform for building, training and deploying models that can run in many environments, from servers to browsers and mobile devices. It includes high-level Keras APIs, tutorials and production-tested tools across the ML workflow, plus ecosystem libraries such as TensorFlow.js for running models client-side in the browser, TensorFlow GNN for graph neural networks on relational data, and TensorFlow Agents for reinforcement learning.

• TensorFlow Fairness Indicators Claim

  Fairness Indicators is an open-source library from the TensorFlow team for evaluating, improving and comparing machine learning models for fairness concerns. It computes commonly used fairness metrics for binary and multiclass classifiers, evaluates dataset distributions, and reports model performance sliced across defined groups of users, with confidence intervals and evaluation at multiple thresholds, and is designed to scale to large datasets and models. The pip package bundles TensorFlow Data Validation, TensorFlow Model Analysis, the Fairness Indicators extension and the What-If Tool for interactive probing.

• Tesseract OCR Claim

  Tesseract is an open-source OCR engine, released under the Apache 2.0 licence, consisting of the libtesseract library and a tesseract command-line program. Since version 4 it includes an LSTM neural network engine focused on line recognition, while still supporting the legacy character-pattern engine. It recognises more than 100 languages out of the box, reads PNG, JPEG and TIFF images, and outputs plain text, hOCR, PDF, TSV, ALTO and PAGE formats; it has no built-in GUI. Tesseract was originally developed at Hewlett-Packard, open-sourced in 2005 and developed by Google from 2006 to August 2017; it is now maintained by an open-source community led by Stefan Weil and Zdenko Podobny.

• Veo Claim

  Veo is Google DeepMind's video generation model, with Veo 3.1 the current version. It turns text prompts and reference images into short video clips and generates the audio natively alongside the picture, including sound effects, ambient noise and spoken dialogue. Google highlights realistic physics, close adherence to prompts and creative control over consistency across shots. Veo is available to try in the Gemini app and in Google Flow, Google's filmmaking tool, and developers can build with it through the Gemini API and Vertex AI. Google aims it at filmmakers, storytellers and creative teams who want to draft or produce video from written direction.

• Vertex AI Vector Search Claim

  Vector Search is Google Cloud's managed vector search engine, now documented as part of Gemini Enterprise Agent Platform (formerly Vertex AI). It is built on Google Research's ScaNN algorithm, the same technology behind Google Search, YouTube and Google Play, and is used to build semantic search, recommendation systems and generative AI applications such as retrieval for large language models. Google positions it as an enterprise-grade service for searching massive datasets with low latency at global scale, including multimodal search.

• Visual Inspection AI Claim

  Visual Inspection AI is Google Cloud's defect-detection service for manufacturers. It finds, classifies and locates defects on production lines, with Anomaly, Assembly and Cosmetic inspection solutions. Google lists it as generally available on a restricted basis through Google sales, priced per camera stream per solution each month, with model training billed per node hour.

Why complete verification

Completing verification can improve the quality of the public evidence available on this product profile. Useful verified information can deepen the profile and make public questions about the named product more answerable. It does not buy directory ranking, review stars, a higher Assurance Score, a favourable certificate outcome, or a guaranteed Google ranking.

Payment remains independent from rankings and assessment outcomes. Start an assessment · Methodology