Daily News · 2 min read

Google AI Updates: October 22, 2026

google google-cloud openai model-garden agents-cli langchain adk google-research healthcare

1. Google Cloud Adds OpenAI’s GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna to Model Garden in Private Preview

Google. Google Cloud is bringing three OpenAI models to Model Garden: GPT-6 Astra for demanding coding, computer use, and research work, GPT-6.1 Sol as a lower-cost option near Astra’s level, and GPT-6 Luna for fast, cheap everyday workloads. Access starts as a private preview for select enterprise customers, and customers can also point local Codex CLI and Desktop workflows at the Gemini platform for inference, with Gemini for business and Google Antigravity support listed as coming soon. Billing runs through Google Cloud Marketplace, so usage draws down existing cloud commitments and counts toward Flexible Savings Plans without a separate OpenAI contract. The models run on dedicated Google Cloud infrastructure with VPC Service Controls and unified audit logging; Zero Data Retention and Confidential Compute options are not yet available. Source

2. Google’s Agents CLI Gains an Extension System and a First-Party LangChain Template

Google. Agents CLI now supports extensions (since v1.5.0): Git-hosted directories with an agents-cli-extension.yaml manifest that can add commands or override built-in subcommands without forking the tool, pinned to a commit SHA via a lockfile. Version 1.8.0 ships a first-party LangChain extension that scaffolds LangChain agents, LangGraph graphs, and Deep Agents with no google-adk dependency and deploys them to Cloud Run through the same path as ADK projects. The extension overrides playground, run, eval generate, and publish gemini-enterprise, but refuses ADK-only commands such as eval dataset and eval optimize. Google labels the extension system experimental, so the manifest schema may still change in breaking ways. Source

3. Google Research’s Blind-Sweep Ultrasound Model Matches Sonographers on Gestational Age

Google. Google Research reported results from a prospective multi-center study, published in JAMA Network Open, of an AI model that estimates gestational age from “blind sweep” ultrasounds, which people without sonography training can perform with low-cost probes. Across 385 participants in Chicago (Northwestern Medicine) and Nairobi (Jacaranda Health), the model reached a mean absolute error of 4.18 days compared with 4.48 days for expert sonographer biometry, meeting the non-inferiority bar. It ran on new hardware (a Clarius C3 probe), and the Nairobi site needed no local fine-tuning. No release of the model, code, or data was announced. Source