Hello from The AI Night,
Today in AI:
Google Launches Project Genie: Create Explorable AI Worlds
xAI Unveils Grok Imagine API With Top Benchmark Scores
Google Brings Gemini 3 to Chrome With Auto Browse
Google DeepMind
Google Launches Project Genie: Create Explorable AI Worlds
Image Source: Google blog
Here's the deal: Google DeepMind launched Project Genie, an experimental web app that lets users create and explore AI-generated interactive environments in real time. Access is rolling out now to Google AI Ultra subscribers in the U.S. (18+).
The Breakdown:
Powered by Genie 3 (world model), Nano Banana Pro (image editing) and Gemini
Three core modes: world sketching from text/images, real time exploration with dynamic path generation and remixing existing worlds
Genie 3 generates environments as you move simulating physics and interactions on the fly
Current limits: 60 second generations, imperfect physics, occasional control latency
Some previewed features (like promptable world changing events) not yet included
First person and third person perspective options available
The bigger picture: This is Google's first public facing world model prototype, a meaningful step toward AI systems that simulate dynamic, navigable environments rather than static outputs. For researchers and creators, it signals where generative media is heading, from images to explorable spaces. The 60 second cap and fidelity gaps show it's early, but the architecture is now testable at scale.
Here's the deal: xAI released Grok Imagine API, a unified video generation and editing API that ranks #1 on Artificial Analysis “text to video” benchmarks while undercutting competitors on both latency and price.
The Breakdown:
ops Artificial Analysis rankings ahead of Google's Veo 3.1, Veo 3 and OpenAI's Sora 2 Pro on quality to cost and quality to latency metrics
Video editing outperforms Kling o1 (57% vs 43%) and Runway Aleph (64.1% vs 35.9%) in human evaluations on instruction following and consistency
Supports text to video, image to video and editing workflows including object add/remove, scene restyling and motion control
Native video audio generation in a single model
Benchmarks measured at 720p, 8 second clips, tested January 28, 2026
The bigger picture: This targets developers who want capable agentic models without the compute costs of full size flagships. A 3B active parameter footprint means practical local inference on consumer hardware while the MoE architecture preserves performance useful for anyone building coding agents or experimenting with tool use workflows on a budget.
Image Source: Google blog
Here's the deal: Google is integrating Gemini 3 directly into Chrome with "auto browse" an agentic feature that handles multi step web tasks autonomously. Available now for AI Pro and Ultra subscribers in the U.S. on MacOS, Windows and Chromebook Plus.
The Breakdown:
Auto browse executes complex workflows: booking travel, filling forms, comparing prices across sites, applying discount codes and managing subscriptions
New persistent side panel keeps Gemini accessible across all tabs for multitasking
Connected Apps integration links Gmail, Calendar, YouTube, Maps, Google Shopping and Flights for cross app context
Supports Universal Commerce Protocol (UCP) a new open standard co-developed with Shopify, Etsy, Wayfair and Target for agentic commerce
Security guardrails require explicit user confirmation for purchases, social posts, and sensitive actions
Personal Intelligence (cross session memory) coming in the following months
The bigger picture: Google is positioning Chrome as the default agentic interface for the consumer web. The UCP standard signals a coordinated push to make browser based AI agents interoperable with major e-commerce platforms a significant infrastructure shift.
What else you need to know:
Google launched AI Plus in 35 new countries bundling Gemini 3 Pro access, Veo 3.1 video generation, expanded NotebookLM features and 200 GB storage shareable with up to five family members.
OpenAI will retire GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini from ChatGPT on February 13 after usage shifted to GPT-5.2 with only 0.1% of users still opting GPT-4o daily.
Microsoft launched Agent Mode in Excel, enabling Copilot to autonomously execute complex multi-step tasks like building reports and reshaping data, with support for both OpenAI and Anthropic models via a model picker.
Anthropic added plugin support to Cowork, letting users bundle skills, connectors, slash commands and sub-agents to customize Claude for specific roles, with 11 open-source plugins available at launch.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
If this was useful, you’ll get the same signal here tomorrow.