Google AI Updates: November 11, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Google Releases Gemini 4 Carbon, Its Strongest Model, at $2 and $10 per Million Tokens Rising to $4 and $20 on November 30
Google DeepMind / Gemini. Google made Gemini 4 Carbon (gemini-4-carbon) generally available on November 10, positioning it above Gemini 4 Argon for frontier cybersecurity, long-horizon agentic workflows, software engineering and scientific reasoning. Google says Argon remains the better fit for most software engineering and knowledge work, while Carbon spends more tokens exploring problems and running verification loops. Developers can use it on the Gemini API paid tier in Google AI Studio and Android Studio; Google AI Ultra subscribers get it in the Gemini app, AI Mode in Search and Google Antigravity. Introductory pricing matches Argon at $2 per million input tokens and $10 per million output tokens, then rises to $4 and $20 from November 30, 2026. Google reports first place on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, 92% on a 375-task agentic honesty benchmark versus 70% for Argon, and a per-attempt breach rate of 0.8% in simulated cyberattack tests versus 3.1% for Argon. The post also introduces /swarm in Antigravity (preview), which turns one prompt into a coordinated team of agents. The post does not state a context window. Source
2. Google Cloud Launches Gemini for Healthcare in US Preview With Skills, EHR and Claims Connectors
Google Cloud / Gemini. Google Cloud introduced Gemini for healthcare on November 10, a preview available in the United States that extends its universal Gemini agent into an orchestration layer for clinical, financial and administrative workflows. It has four parts: reusable skills (for example medical records synthesis, pre-appointment preparation and payer benefit determination), connectors to EHR, claims and care-management systems that preserve existing access controls, an open partner ecosystem, and a governed platform with HIPAA-oriented controls, audit logging and citations. Named connectors include Epic and Oracle Health (read-only), InterSystems FHIR Server and Optum eligibility services, the latter two in limited release. Launch customers include HCA Healthcare, Highmark Health and Northwell Health, whose statements on the page are early-stage rather than measured results. Google states that clinical data, prompts and outputs are not used to train its foundation models. The post gives no pricing or general availability date. Source
3. DeepMind’s AlphaProof Code Uses Gemini Carbon Agents to Formally Verify the Signal Protocol Implementation in About Four Weeks
Google DeepMind. Google DeepMind introduced AlphaProof Code on November 10, an agentic framework for formal verification in which humans decide which properties to prove and AI agents write the proofs. Applied to the implementation of the Signal protocol, which also protects messages in Google Messages, agents powered by Gemini Carbon produced more than six million lines of Lean proofs. Human experts reviewed each formal specification, and the agents flagged about 1% of the codebase for human review while Lean checked the rest. DeepMind says the work took about four weeks, where expert teams historically needed years, and describes it as the first formal verification of a widely deployed cryptographic protocol’s implementation at this depth. In a test, the team reintroduced two previously fixed bugs into an isolated copy of the code and the system produced machine-checked proofs that the buggy versions were insecure. The verification artifacts are being open-sourced, and a technical paper titled “Formal Verification of Signal” accompanies the post. Source
4. Gemini Robotics 2.5 Reports 30% to 100% Higher Success Rates Across More Than 200 Tasks
Google DeepMind / Robotics. Google DeepMind published Gemini Robotics 2.5, which it calls its most capable vision-language-action model to date, aimed at tool use, deformable objects and delicate items such as serving food with tongs, tying a bow and pipetting. Across more than 200 tasks, DeepMind reports 30% to 100% higher success rates than previous baselines depending on task type, with tasks completed 20% to 30% faster. In a proof of concept on Mercedes-Benz production lines, Apptronik’s Apollo 2 humanoid reached a 96% success rate on automotive kitting. Safe task completion rose 33% around obstacles and 43% around nearby people, and Gemini 4 Argon reached 99.5% on jailbreak resistance against 82% for Gemini Robotics ER 2. DeepMind also released an open-source benchmark, ASIMOV-Jailbreak, for robotics-specific adversarial prompts. The model is rolling out to DeepMind’s robotics partners rather than broadly. Source
5. BigQuery Measures Reaches GA, Defining BI Metrics on BigQuery Graph for Agents and Dashboards
Google Cloud / BigQuery. Google Cloud made measures in BigQuery generally available, letting teams define governed business metrics directly on BigQuery Graph entities so that dashboards and AI agents use the same definitions and relationship context. Measures are grain-locked, which avoids overcounting an account once per joined transaction row, and can be queried with SQL using GRAPH_EXPAND and AGG() or with GQL. The GA release adds composite and derived measures validated at schema compile time, multi-fact queries through a start_node argument that avoids inflated cross-fact joins, and metadata options such as descriptions and synonyms. Graph and measure metadata is ingested into Knowledge Catalog, and the feature integrates with Data Studio, Looker (through LookML-based derived models), Conversational Analytics and Data Agent Kit. The post gives no performance benchmarks. Source
6. Google Cloud Publishes a Reference Architecture for Running CodeMender Scan, Verify and Fix in Pull Requests
Google Cloud / Security. Google Cloud authors described how to wire CodeMender, part of Google’s AI Threat Defense platform, into Git pull requests through Cloud Run, with an open reference implementation called codemender-gitops. A pull request triggers a scan that posts findings, and reviewers reply with /verify, /fix and /approve commands to run a sandboxed exploit attempt, generate a patch and commit it. The post does not state a general availability or preview status for CodeMender itself. The security design uses GitHub OIDC with Workload Identity Federation instead of stored credentials, one Cloud Run job per exploit in a second-generation microVM with a dedicated service account and no network egress, and marks incomplete verification runs as inconclusive rather than safe. The security/codemender-gate commit status only blocks merges when branch protection is configured. Source
7. Google Cloud’s Gemini Factory Program Helps ISVs Build MCP Connectors, Skills and A2A Agents for Gemini Enterprise
Google Cloud / Agents. Google Cloud described the Gemini factory (previously called the Agent Factory), a co-innovation program that helps enterprise software partners move from a concept to agentic solutions listed on Google Cloud Marketplace. Partners choose among four primitives: MCP connectors registered in the Agent Registry, declarative .md skills in the Skills Registry, plugins that bundle both, and standalone A2A agents that run in the partner’s own environment. Tooling includes Antigravity, prebuilt Rapid Agent Development skills, the Agent Development Kit and the A2A protocol. Qualified solutions can receive seed credits at Marketplace publishing and performance credits for inference costs, though the post gives no dollar amounts. Workshops run in San Francisco, New York City, Sunnyvale, London, Paris, Tel Aviv and Tokyo, and nine partner agents are featured, with partner-reported prototype times as short as under 24 hours. Source
8. Gemini Intelligence Adds Rambler, Create My Widget and Device Assistance, and Reaches Pixel 10 and Motorola Devices
Google / Android. Google announced new Gemini Intelligence features and devices on November 10. With the November Pixel Drop, the Pixel 10, Pixel 10 Pro XL and Pixel 10 Pro Fold receive the latest features, and the motorola signature 27, Motorola’s first Gemini Intelligence device, opens for US pre-orders on November 19. Rambler, a Gboard feature that turns spoken thoughts into organized text, comes to the Pixel 10 family and the motorola signature 27. Task automation for actions such as ordering food and booking rides across apps expands to the UK, India and Japan. Create My Widget builds a homescreen widget from a text description using Gmail, Calendar and Search and rolls out over the next few weeks, and Device Assistance handles tasks such as clearing storage on select devices running Android 17 or later. Proactive suggestions on Android Auto are described as coming soon. Availability varies by device, region and language. Source