The AI landscape is accelerating faster than ever and this week, it’s all about smarter browsing, seamless coding, and real-time video generation.

🌐 ChatGPT Atlas by OpenAI - The future of browsing begins here. Atlas is a new macOS browser with ChatGPT built directly into it, offering real-time assistance anywhere on the web. From summarizing pages to booking events or writing inline, Atlas combines productivity and personalization with memory, smart search, and on-page AI help now available globally across all ChatGPT plans.

💻 Claude Code on the Web - Anthropic brings AI-powered coding to your browser. This beta research tool lets developers assign, monitor, and manage multiple coding tasks in real time directly from the cloud. With GitHub integration, parallel task execution, and secure sandboxed environments, it’s redefining how teams ship code faster and safer.

🎬 Krea Realtime Video Model - Creativity meets speed. Hosted on Hugging Face, this 14B parameter model generates and edits video in real time achieving up to 11 frames per second on a single GPU. Whether restyling mid-generation or streaming webcam-to-video transformations, Krea sets a new benchmark for interactive generative video.

From intelligent browsers to collaborative coding and live AI video synthesis, this week’s breakthroughs are reshaping how we browse, build, and create in real time.

The Future of Browsing Begins with Atlas

OpenAI has introduced ChatGPT Atlas, a new macOS browser with ChatGPT built directly into it. Atlas transforms browsing by letting ChatGPT assist anywhere on the web—summarizing pages, writing text, conducting research, and even performing actions like booking events or ordering groceries through its agent mode. It features built-in memory for personalized context, so ChatGPT remembers relevant information from previous sessions while giving users full control over what’s stored. Atlas also enhances productivity with tools like inline writing help, smarter search results, and a constantly accessible sidebar for on-page assistance. The browser is available now for Free, Plus, Pro, and Go users globally, with beta access for Business and Enterprise accounts. Support for Windows, iOS, and Android is coming soon.​

Claude Code on the Web: AI-Powered Coding Made Easy

Anthropic has launched Claude Code on the web, a new beta research preview that lets developers delegate multiple coding tasks directly from their browsers. With cloud-based sessions running on secure, isolated environments, users can connect GitHub repositories, assign tasks, track real-time progress, and steer work dynamically. Claude Code on the web supports parallel task execution across repositories to speed up bug fixes, routine tasks, and backend changes with test-driven development. Available now for Pro and Max users, the browser tool emphasizes security with sandboxing and network restrictions, and also features a mobile version on iOS for coding on the move. This innovation streamlines workflows and accelerates software delivery without the need to open terminals or local environments.

Krea Real-Time AI Video Generation Model

Krea Realtime Video is an advanced AI-powered video generation model hosted on Hugging Face, featuring 14 billion parameters and real-time capabilities. Distilled from the Wan 2.1 14B text-to-video model using the Self-Forcing technique, it achieves impressive text-to-video inference speeds of 11 frames per second on a single NVIDIA B200 GPU. The model supports real-time interactive video editing, allowing users to modify prompts mid-generation, restyle videos on the fly, and see initial frames within one second. It also enables streaming video-to-video transformations by processing webcam inputs or canvas sketches for controllable video synthesis and editing. Designed for creative exploration and rapid iteration, Krea Realtime Video offers both streaming and batch sampling modes, leveraging cutting-edge optimizations in memory and attention mechanisms. This open-source solution is accessible via Hugging Face with inference code and integrates seamlessly with the diffusers library for enhanced modular video generation workflows.​

Hand Picked Video

In this video, we’ll look at how Google’s Gemini Nano Banana model edits complex images, replacing objects, improving vibes, restoring photos, and even making creative transformations.

Top AI Products from this week

• ProblemHunt - Problem: the main reason startups fail is a lack of market need. 42% of startups built solutions that didn't solve real problems. Solution: we manually find people with unresolved problems they are willing to pay to solve.

• Manus 1.5 - Manus 1.5 is a faster, smarter AI agent system for research, analysis, and full-stack web app creation. It delivers 4× speed, higher reliability, deeper reasoning, collaboration, and a new Library—plus seamless, production-ready app building through chat.

• Ito - Ito is an open source voice assistant for Mac and Windows that transforms your intent into smart text in any app. Speak naturally to write emails, messages, or code without typing. Say intent, not just words.

• Metorial - Build AI agents with 600+ integrations in hours, not months. Metorial's MCP-powered platform handles OAuth, monitoring, and deployment automatically. Open-source with Python & TypeScript SDKs. Easily self-hostable.

• Claude Skills Hub - A curated directory of Claude Code skills for everyone. Browse, discover, and download skills to enhance your Claude experience.

• Cheers GEO - When someone asks ChatGPT for a local service, AI recommends based on reputation, not ads. Cheers optimizes your online presence and turns your employees into a review engine so that AI recommends your business first.

This week in AI

• Claude for Life Sciences - AI boosts research by integrating with lab platforms like Benchling and PubMed, speeding drug discovery, protocol writing, and regulatory tasks.

• Gemini API Integrates Google Maps - Google's Gemini API now integrates Google Maps data, enabling AI apps to deliver real-time, location-aware responses with interactive map widgets and precise local info.

• AI Content Control - Pinterest adds new tools letting users limit AI-generated content in feeds across categories like art and fashion, with enhanced AI labels for transparency and user control.

• DeepSeek-OCR Advanced Visual Text Compression - DeepSeek-OCR compresses visual text into efficient tokens, enabling high-fidelity OCR and markdown document conversion with minimal compute and strong accuracy.

• Yelp's AI Host & Receptionist - Yelp’s new AI-powered Host and Receptionist answer calls, take reservations, and handle customer inquiries for restaurants and local businesses 24/7, enhancing service and efficiency.

• Fish Speech Free TTS Model - Fish Speech is an open-source text-to-speech framework leveraging large language models and dual autoregressive architecture to deliver high-quality, multilingual speech synthesis with voice cloning and fine-tuning support.

Paper of The Day

The paper introduces Visual Attention Reasoning (VAR), a novel framework for multimodal large language models that improves reasoning by using structured search with backtracking. VAR decomposes the reasoning process into traceable evidence grounding and search-based chain-of-thought generation, allowing self-correction through backtracking. Guided by a multi-faceted reward system with semantic and geometric self-verification, VAR mitigates hallucinations and enhances accuracy. Experimental results show VAR-7B sets new state-of-the-art in hallucination and safety benchmarks, outperforming leading open-source and some proprietary models.

To read the whole paper 👉️ here