Google AI Updates: October 29, 2026

1. Google DeepMind Announces Gemini 4 Carbon, Its New Frontier Model, With an Initial Release to Cyber Defenders

Google announced Gemini 4 Carbon, which it calls its strongest model, focused on cybersecurity and long-horizon agentic tasks. It arrives two weeks after Gemini 4 Argon. Google reports a top score of 59 on the Artificial Analysis Intelligence Index v5.0, 88.2% on SEC-bench Pro for vulnerability finding, 54.8% on ExploitGym, 71.0% on CWE-bench patching, and 1768 Elo (#4) on WebDev Arena. The company also reports a 0.0% attack success rate on Gray Swan’s indirect prompt injection benchmark (Argon: 0.7%) and nearly 80% fewer high-severity misalignment alerts in internal coding environments. Carbon is going first to a subset of cyber defenders, then to trusted partners through the Fairwind Program, with broader developer and consumer access to follow. Google is also taking part in the U.S. government’s voluntary pre-release model testing. Pricing is $2/$10 per million input/output tokens during an introductory period, rising to $4/$20 afterward, with cached input tokens 95% off. Source

2. Android Opens to Third-Party Coding Agents and Adds APIs for Agents to Act Inside Apps

Google used its Android Dev Summit to add agent support across Android development and apps. Android Studio Rabbit 2 adds “Bring Your Own Agent,” which uses the open Agent Client Protocol to run more than 40 coding agents, including Anthropic’s Claude Agent, OpenAI’s Codex, and Google Antigravity, with access to project structure, the emulator, and Logcat. The Android CLI is now an official agent plugin for Claude Code, Codex, and Antigravity. Google says its new terminal Compose previews let agents finish UI tasks 77% faster with up to 3x fewer tokens. On the app side, the AppFunctions API (beta) uses MCP so apps can expose functions that on-device or system-wide agents can call. The ML Kit GenAI Agentic API (alpha, Gemini Nano 4) runs the agent loop on-device, and A2UI for Jetpack Compose (alpha) lets agents render app-native UI. Google also launched Android Bench 2.0, more than 30 Android Skills packs, and the XREAL AURA Android XR glasses ($1,099, shipping mid-November). Source

3. Google Launches the Connected Apps Platform to Bring Third-Party Services Into Gemini and Search

Google introduced the Connected Apps Platform, which lets any business submit an app that users can link and act through in the Gemini app, Google Search, or both. Example actions include designing a flyer in Canva and building a playlist in YouTube Music. Developers integrate through a cloud-hosted MCP server or Android MCP with the AppFunctions APIs, then register, test, and submit apps in the Google Connected Console, where they can monitor traffic, latency, and error rates. Gemini already has more than 50 connected apps, including Adobe, Experian, and Peloton, and these are coming to Search in the next few weeks. The platform is in beta for U.S. developers, with a waitlist-based public preview. Source

4. Chrome’s Enhanced Autofill Pulls Order Numbers and Flight Bookings From Gmail

Google began rolling out enhanced autofill in Chrome, which uses Gemini with Personal Intelligence to suggest details from Gmail, such as order numbers, tracking information, and flight bookings, when users fill out web forms. The feature is for Gemini in Chrome users who have Personal Intelligence enabled, and it can be turned on or off, with access controlled in Chrome settings. Google says sensitive data is encrypted. The rollout is starting in the U.S. on desktop, Android, and iOS. Source

5. Oracle Becomes the First Outside Cloud to Offer Google TPUs

Alphabet said Oracle will offer Google’s TPUs, the first time customers can access TPUs outside Google’s own infrastructure. Analyst estimates of the opportunity vary widely: TD Cowen projects $9.4 billion for the second half of 2026 and $62.5 billion for 2027, while Wells Fargo estimates $99 billion next year. Source

6. Alphabet Misses Q3 Estimates as AI Spending Pushes Free Cash Flow Negative

Alphabet missed third-quarter earnings and revenue estimates, and its shares fell almost 10% after hours. Free cash flow was negative by more than $9 billion. The company held its full-year capital expenditure guidance at $195 billion to $205 billion, mostly for AI infrastructure. Other Bets, which include Waymo and Isomorphic Labs, had $372 million in revenue and a $2.05 billion operating loss, compared with a $1.43 billion loss a year earlier. Source

7. LiteRT in Google Play Services Adds GPU Inference Through the CompiledModel API

Google added GPU inference to LiteRT in Google Play services, Android’s shared on-device inference runtime, through the new CompiledModel API. Google says GPU inference runs about 1.4x faster than the legacy GPU delegate, and apps that moved to the Play services runtime were about 8 MB smaller in its testing, since the runtime ships with Play services rather than inside each app. Support covers phones back to Android 7 (API 24), Wear OS 4+, and Android Automotive OS 11+, starting with play-services-litert:17.0.0. The legacy GPU and custom delegate APIs are deprecated, NNAPI support will be removed starting in 2027, and NPU support is in development. Source