Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 6 min read

Google AI Updates: November 12, 2026

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

1. Google Launches the Gemini Decisions API and Gemini Decision 1 for Closed-Set Decisions at $0.10 per Million Input Tokens

Google released the Gemini Decisions API on November 11, a new endpoint backed by the gemini-decision-1 model that answers bounded questions instead of generating text. A request carries one multimodal input (text, images, audio, video or PDFs, up to 256K tokens) and up to 1,000 independent questions of three types: yes/no, multiple choice over 2 to 255 options, and scoring against an ordered rubric. Each answer returns probabilities and a confidence score from 0 to 1, computed in a single forward pass with no token-by-token decoding. Google cites 150 to 250 ms end-to-end latency for contexts up to 16K tokens, roughly flat from 1 to 100 questions per request. Pricing is $0.10 per million input tokens with no output charge, and the shared input is billed once per request. The API is available through the Gemini API and Google AI Studio with a no-cost tier, with Google Cloud support listed as coming soon. Benchmark comparisons are published as charts only, so exact scores are [NEEDS VERIFICATION], and the post does not state whether the API is in preview or generally available. Source | Changelog

2. DeepMind’s AlphaGenome 2 Conditions DNA Predictions on a Single Cell Type and Nearly Doubles Causal Variant Identification

Google DeepMind introduced AlphaGenome 2 on November 11, a DNA sequence model that predicts how sequences and variants affect gene regulation in a specific tissue, cell type or cell state. The original AlphaGenome emitted a fixed set of outputs for many tissues in parallel; the new model takes the condition as an input, and a cell encoder lets researchers derive that condition from their own single-cell data, including for cell states absent from training. It reads about one million base pairs, adds DNA methylation to its readouts, and was trained on human and mouse bulk and single-cell data including 107 million single-cell profiles across 68 tissues. On the TenK10K benchmark of 28 primary immune cell types it identified 30% of causal variants against 16% for the original model fine-tuned on single-cell data, and 43% of directional gene effects against 23%. The model is available for non-commercial research through the AlphaGenome API, with commercial research access in private preview. The post does not mention a weights release or a journal for the linked paper. Source

3. AlloyDB AI Adds ai.classify and ai.extract SQL Functions in Preview

Google Cloud announced two functions in the AlloyDB AI query engine that run text classification and structured extraction with Gemini models directly from PostgreSQL. ai.classify assigns single or multiple labels from user-defined categories, with optional category descriptions and few-shot examples, and ai.extract returns JSON matching a JSON schema or a plain-text field list. Both accept single rows, arrays and cursors. The preview requires PostgreSQL 18, version 1.6.1 or later of the google_ml_integration extension, and SET google_ml_integration.enable_preview_ai_functions = TRUE;. When the functions appear in a WHERE clause, smart batching (previously called AI Function Acceleration) groups rows into micro-batches before calling the model. The post gives no performance numbers, specific model versions or pricing. Source

4. Google Labs Opens a Waitlist for Tabletop, a Shared Space That Generates Group Planning Tools From Prompts

Google Labs announced Tabletop on November 11, an experiment for real-time group collaboration. Users add links, files and ideas to a shared space, then prompt @tabletop or start from a template to generate interactive tools such as trip itineraries, polls, expense splitters and tournament brackets. Spaces are shared by invite link and edited by several people at once. Access is by waitlist at labs.google/tabletop for US users aged 18 and over. The post does not say which model powers it. Source

5. Google Research Describes a Wearable World Model That Simulates 12-Week Habit Changes for Health Coaching

Google Research published work on November 11 on a “coachable health age” computed from smartwatch signals (resting heart rate, steps, sleep duration and Active Zone Minutes) and designed to respond to behavior change. In the All of Us cohort (n = 22,748), participants whose health age exceeded their calendar age at baseline were more likely to develop hypertension, cardiovascular disease or type 2 diabetes over eight years. SensorWM, an action-conditioned world model pre-trained on about 2.28 million person-weeks from roughly 100,000 consenting Fitbit users, forecasts 20 physiological markers over 12 weeks from two weeks of baseline data and is reported to outperform TimesFM. The system simulates 125 habit combinations per person, filters them for feasibility, and has a language model write a 12-week plan. Certified coaches preferred the simulation-grounded plan in 71.4% of paired reviews, and physicians rated 99.2% of plans within safe exercise intensity limits. The work is described in two arXiv preprints and as an investigational prototype evaluated offline, with no product announced. Source

6. Google Cloud Introduces a Professional Agentic Architect Certification With a Hands-On Lab Component

Google Cloud introduced the Professional Agentic Architect certification as part of its Gemini Agent Ready (GEAR) initiative, which it says nearly one million developers are using. The exam has two parts: a proctored multiple-choice section on system design, cost and security governance, and hands-on labs on Google Skills in which candidates build, test and debug agentic workflows in a live cloud environment. It covers five domains: low-code agent building, coding agents for application development, custom agents on the Gemini platform with open source frameworks, evaluation and deployment, and security and governance. According to the post, the program started as internal training for forward deployed engineers and partners over the past four months. Exam cost, length, languages and prerequisites are not stated in the post. Source