AI News of the Weeks (19th December)
TLDR: Key AI Developments This Week Google launched Gemini 3 Flash as the default model for AI Mode in Search, delivering Pro-grade reasoning at Flash-level speed, while NVIDIA released the Nemotron 3 family for agentic workflows and synthetic data generation. Meanwhile, OpenAI launched the ChatGPT App Store for developers to build and monetize applications, Zara integrated AI into retail production for faster content generation, and ChatGPT rolled out global group chats for collaborative AI interactions.
Gemini 3 Flash Becomes Google’s Default Fast Reasoning Model
Google has officially launched Gemini 3 Flash globally, making it the default model for AI Mode in Search and rolling it out across Google AI Studio, Vertex AI, and the new Antigravity developer platform. Unlike earlier trade-offs between speed and intelligence, Gemini 3 Flash delivers “Pro-grade” reasoning at Flash-level latency and cost, powered by a new Dynamic Thinking Modulation system that adjusts reasoning depth per query. It runs 3× faster than Gemini 2.5 Pro, uses 30% fewer tokens, supports code execution on visual inputs, and has become Google’s top software engineering model, scoring 78% on SWE-bench Verified, even outperforming Gemini 3 Pro. This unlocks real-time, autonomous agent loops, large-context analysis (up to 1M tokens) without heavy RAG infrastructure, and more interactive generative experiences in Search.
NVIDIA Releases Nemotron 3 for Agent-First AI Systems
NVIDIA has released the Nemotron 3 family, Nano (30B), Super (100B), and Ultra (500B), open models purpose-built for agentic workflows and synthetic data generation. Unlike general-purpose models that struggled with long-horizon planning and required massive hardware, Nemotron 3 uses a hybrid Mixture-of-Experts architecture that activates only a fraction of parameters per token, allowing the 30B Nano model to deliver 70B-class performance on a single workstation GPU. NVIDIA also released model weights, curated datasets, and the NeMo Gym RL environment, enabling local multi-agent “swarms,” privacy-first deployments, and cost-effective self-hosting. This marks a major shift toward high-performance, offline-capable agent systems without reliance on cloud APIs. Read more
OpenAI Launches the ChatGPT App Store
OpenAI has officially launched the ChatGPT App Store, allowing developers to build, publish, and monetize full-featured applications directly inside ChatGPT. Moving beyond simple chatbots, the platform enables vertical tools, automated workflows, analytics dashboards, and specialized utilities to be distributed through a native marketplace embedded in the ChatGPT tools menu. For developers, this introduces standardized discovery, lower adoption friction, and direct monetization paths. For clients, it turns ChatGPT into a centralized platform for installing task-specific AI solutions, reducing integration overhead and positioning ChatGPT as an operating system for AI-powered work. Read more
Zara Shows How AI Quietly Reshapes Retail Production
Zara is integrating generative AI into its retail workflows to generate new product imagery by creating variations from existing photoshoots, rather than starting each shoot from scratch. Instead of replacing creative judgment, the system targets high-frequency, repeatable tasks, allowing AI to adapt imagery across regions, channels, and campaigns while keeping human models, consent, creative direction, and quality control intact. Embedded directly into Zara’s production pipeline and data infrastructure, this approach reduces costs, accelerates content turnaround, and supports the fast-fashion model by compressing production cycles and improving feedback loops between inventory, demand, and customer response. Read more
ChatGPT Rolls Out Global Group Chats
OpenAI has launched Group Chats globally, allowing up to 20 users to collaborate with ChatGPT in a shared conversation. This transforms ChatGPT from a one-on-one assistant into a collaborative team member, solving prior issues with siloed interactions and fragmented context. The AI introduces “smart silence,” responding only when mentioned or contextually relevant, while maintaining a unified context window shared across all participants. With profile awareness, privacy isolation from personal chats, and cross-plan compatibility, Group Chats enable real-time brainstorming, live debugging, and collaborative planning, bringing AI directly into group workflows instead of forcing copy-paste coordination. Read more
