Friday, November 20, 2026 Edition No. 244
Search
Daily News Fully automated · 6 min read

Google AI Updates: November 19, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Gemini 4 Neon Launches With $3 and $15 Per Million Token Pricing and a 1M-Token Context Window

Google DeepMind released Gemini 4 Neon, a model for complex software engineering, knowledge work and science that sits between Carbon (long-horizon projects) and Hydrogen (fast everyday execution) in the Gemini 4 family. It is available in the Gemini API as gemini-4-neon, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app model dropdown for Pro and Ultra subscribers, with pricing of $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, four thinking levels (high by default, plus xhigh) and a 1M-token context window. Google reports that it ranks #1 on the Artificial Analysis Coding Agent Index (58.6) and #3 on the Artificial Analysis Intelligence Index (61 at xhigh), reaches 65.2% on Terminal-Bench 4.1, and costs 32% less per task than Carbon at the high setting, and a cyber variant is rolling out to trusted defenders in the Fairwind Program. Source

2. Genie 4 Arrives in Project Genie With Real-Time Native Audio and Mid-Session Prompting

Google DeepMind made Genie 4 available in a new Project Genie experience, describing it as its most realistic and interactive world model so far and the first Genie model to pair visuals with real-time native audio. Users can type new prompts mid-session to change the weather, add characters or redirect the action, and Genie 3 remains in the model picker with sessions extended from 60 seconds to 3 minutes. The release also adds a Project Genie agent that builds worlds from a conversation and a gallery for publishing and remixing worlds; it is rolling out to eligible Google AI Ultra subscribers, with a waitlist for Google AI Pro, and all output carries SynthID watermarks. Source

3. Gemini API Accepts Office Files and Slices Documents, Spreadsheets, Video and Audio

Google added multimodal file processing to the Gemini API, which now accepts Word, PowerPoint and Excel files and their OpenDocument equivalents directly. Developers can also slice documents, spreadsheets, videos and audio in prompts and function responses by page range, worksheet tab, cell range, time interval or content layer, and are billed only for the selected content. The same November 18 changelog lists Gemini 4 Neon (gemini-4-neon) as generally available on the paid tier. Source

4. The Gemini App Comes to Lower-Memory Android Go Devices

Google re-architected the Gemini app to run on eligible Android (Go edition) phones with less memory, which previously offered only basic voice commands. The app brings Gemini Live voice conversations in more than 45 languages, camera tools for learning from or translating notes and forms, and image creation with Nano Banana. It is available from Google Play or by long-pressing the power button on supported phones, and will come preloaded on upcoming devices from partners such as Transsion. Source

5. Google Details Seven Months of Sashiko, Its Open-Source Agentic Code Reviewer for the Linux Kernel

Google published lessons from Sashiko, an open-source agentic code review system built on Gemini and donated to the Linux Foundation, which splits bug finding from bug verification across up to seven parallel specialist sessions and a lead verifier. Google reports 241,000 patch reviews, more than 2,000 mainline commits crediting Sashiko, and a drop in the 60-day post-merge fix rate from 3.4% to 2.2% in Linux 7.2, a result the authors say is too early for strong claims. On an internal benchmark of nearly 1,000 real kernel bugs that passed human review, Sashiko with Gemini 4 Carbon at high thinking caught 75.8%; the service runs at sashiko.dev and the code is on GitHub. Source

6. Open Knowledge Format v0.3 Adds Conformance Rules, Current-Only Verification and Agent Skills

Google Cloud released v0.3 of the Open Knowledge Format (OKF), an open markdown and YAML specification that gives AI agents context, with a formal conformance section, trust tiers that count only verifications made after the last meaningful change, and a superseded_by field for deprecated concepts. The release adds self-describing bundles through frontmatter in the root index.md, relative path resolution, x- extension keys, community use-case guides, and two Agent Skills format skills, okf-generate and okf-consume. The project moved to a standalone GitHub organization, and the version has two breaking changes: path fields resolve relative to their file, and only current verifications count toward trust. Source

7. Google Cloud Outlines an Agentic Stack for Envoy With Token Rate Limits and MCP Routing

Google Cloud described how the open-source Envoy proxy can act as an AI-native data plane through four pillars: a native body parser that normalizes OpenAI, Anthropic and Gemini schemas, a tokenomics engine with tokens-per-minute and tokens-per-day limits, model and MCP routing with KV-cache prefix affinity, and guardrails through ext_proc callouts to Model Armor and ShieldGemma. For MCP traffic, Envoy can route on the Mcp-Method and Mcp-Name headers from the 2026-07-28 MCP specification and reject header and body mismatches with HTTP 400. The post also pairs Envoy with Agent Substrate for scale-to-zero agent sandboxes and egress control; it contains no configuration samples. Source

8. Google Adds Temporary Threads to AI Mode in Search, Keeping Chats Out of History and Personalization

Google introduced temporary threads in AI Mode in Search. Conversations stay out of search history and personalization, and the session is detached from the user’s account. The feature is rolling out globally on desktop and mobile in all supported languages. The announcement does not state how long Google retains the data. Source