This week in AI, creativity, efficiency, and web browsing get major upgrades. From OpenAI’s next-gen video model to a new efficiency breakthrough in LLMs and even an AI-powered browser that works like an agent, here’s what’s making headlines:

🎬 Sora 2: Next-Gen AI Video Model
OpenAI launches Sora 2, a cinematic-ready video and audio model with synchronized dialogue, sound effects, and realistic world simulation, available first in the U.S. and Canada.

⚡ DeepSeek V3.2 Exp
DeepSeek introduces a sparse attention innovation that cuts API costs by up to 50% while handling longer contexts more efficiently, marking a leap in transformer efficiency.

🌐 Opera Neon AI Browser
Opera unveils Neon, an AI-native browser with “Tasks” and “Neon Do” that can autonomously browse, compare, and act across sites—offering privacy-first agentic web workflows.

From smarter video creation to leaner models and AI-native browsing, these breakthroughs show how rapidly AI is shaping the future of work and entertainment.

OpenAI Sora 2 just leaked and shook the AI world

Sora 2 is OpenAI’s latest video and audio generation model, delivering more realistic, physically-accurate, and controllable outputs compared to its predecessor. The new Sora app lets users create, remix, and share videos—featuring synchronized dialogue, sound effects, and “cameos” for inserting likenesses with high fidelity after a quick recording. Built around advanced world simulation capabilities, Sora 2 excels at cinematic and anime styles, models realistic physical dynamics, and injects real-world elements; it’s available first in the U.S. and Canada on iOS, with a web version and pro access via sora.com coming soon. The platform prioritizes community creation, wellbeing tools, and user control over AI-generated content and interaction limits, especially for teens, aiming for a healthier, more creative social experience.

DeepSeek V3.2 Exp Efficient Long Context AI

DeepSeek-V3.2-Exp is an experimental AI model released by DeepSeek AI as an intermediate step toward their next-generation architecture. Building on the previous V3.1-Terminus model, it introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to significantly improve computational efficiency and reduce costs during long-context training and inference, while maintaining nearly identical output quality. Benchmarks show that DeepSeek-V3.2-Exp performs on par with V3.1-Terminus across various reasoning and agentic tool use tasks. The model is open source with inference support across multiple platforms, and the new DSA mechanism marks a major innovation toward more efficient transformer architectures, offering up to 50% reductions in API costs and improved token selection during long document processing. It targets scenarios requiring extended text sequence handling with lower computational overhead, positioning it as a promising advancement in efficient large language models.

Opera Neon AI Browser: Smart Agent for Seamless Web Tasks

Opera Neon is a premium, agentic AI-powered browser designed for power users who extensively use AI in their daily workflows. Unlike traditional browsers that only display web pages, Neon introduces “Tasks,” self-contained workspaces where the AI understands the context and can analyze, compare, and act across multiple sites, documents, and AI chats simultaneously. The browser features “Neon Do,” a function that can autonomously perform complex browsing tasks such as opening and closing tabs, filling forms, comparing data, shopping, booking, or even applying for jobs, all within the user’s browser session locally without sending data to external clouds. Neon also offers “Cards,” reusable prompt instructions that can be customized or downloaded from the community to automate repeated tasks efficiently. Built on Opera’s 30 years of browser expertise, Neon includes essential features like VPN, ad blocker, and bookmarks, while emphasizing user privacy and control. It is available as a subscription service with early access for invited users.

Hand Picked Video

In this video, we’ll look at how GPT‑4o’s brand new image generation capabilities let you create stunning, photorealistic visuals—right from a simple prompt. From beautifully rendered text and diagrams to whimsical illustrations and sleek product mockups, 4o blends deep world knowledge with visual precision. We’ll explore how it handles complex scenes, multi-turn edits, and even integrates image inspiration—all natively inside ChatGPT

Top AI Products from this week

• Everyday - The easiest way to complete tasks across your favorite tools. Describe what you need, and Everyday handles it for you.

• Verdent Deck – Verdent Deck coordinates multiple AI agents to tackle complex coding tasks in parallel. Sessions can run on their own while you step away, collision-free execution, clear insight, and a seamless flow that turns ideas into real, shippable code.

• Ask Brave – Ask Brave is a new interface from Brave that unifies AI chat and web search. Get comprehensive, grounded answers and follow up with questions in a single, privacy-first conversation. Available on any browser.

• Granola Recipes - Recipes are saved prompts written by experts that work with your meeting notes, combining the power of great AI prompts with the nuance of your work conversations.

• MCP by Alloy Automation - MCP by Alloy Automation is an AI developer toolkit and registry that exposes Quickbooks, Xero, Salesforce, Slack and hundreds more as safe tools for AI agents. Start building immediately at ai.runalloy.com.

• Genspark Photo Genius - OpenAI Realtime voice tech meets Nano-Banana image AI. Now you can edit photos just by talking. Tell it what you want, watch magic happen: • Perfect makeup, hair & outfit styling • Rescue your photo fails Just say it, Genspark nails it!

This week in AI

• Claude Agent SDK Overview - Claude Agent SDK lets developers build powerful AI agents with autonomy and control. It supports file operations, web search, code exec, and integrations for versatile workflows.

• Lovable Cloud & AI Launch - Lovable Cloud & AI lets anyone build complex AI-powered apps with backend functions by just prompting. 100k+ new ideas launch daily with seamless AI from Google Gemini.

• Advancing AI Leadership in California with SB 53 - California’s SB 53 law mandates transparency for top AI firms, requiring safety reporting, whistleblower protections, and fostering ethical AI innovation through CalCompute.

• Microsoft 365 Introduces Vibe Working with AI Agents - Microsoft 365 Copilot's new vibe working lets AI agents create and refine Excel & Word files via simple prompts, boosting productivity with human-agent teamwork.

• Ring-1T Preview - Apple’s internal chatbot Veritas tests new Siri AI features like personal data search and photo editing, aiming for a revamped, more reliable Siri by March 2026.

Paper of The Day

Big Survey introduces a large-scale dataset and method to generate structured summaries from thousands of academic sources on a single topic. It includes over 7,000 survey summaries and 430,000 abstracts. The approach uses category-based alignment and sparse transformer (CAST) for efficient, non-redundant summarization of diverse documents, outperforming existing methods. This update highlights how BigSurvey tackles the challenges of multi-document summarization at scale by organizing and processing extensive research content to produce comprehensive summaries. It advances automatic summarization for wide academic literature collections.

To read the whole paper 👉️ here.