Latest AI Developments & Weekly Updates
Last updated: November 14, 2025 | by the VEGA AI Team
This page tracks weekly AI updates across model launches, agent tools, research breakthroughs, and major partnerships. Updated every Friday.
9th Nov – 15th Nov, 2025
Key Releases:
Google Private AI Compute: privacy-preserving cloud to run heavier AI with on-device-like protections. Read More
ElevenLabs Iconic Voice Marketplace: legally license celebrity/historic voices for AI narration and content. Read More
MiniMax Hailuo-2.3 (video): boosts realism in motion, stylization, and character micro-expressions with better command following. Read More
Moonshot K2 Thinking: new “thinking” model reports wins vs GPT-5/Claude on HLE/BrowseComp at far lower cost. Read More
2nd Nov – 8th Nov, 2025
Key Releases:
OpenAI × Amazon ($38B): long-term cloud deal to secure massive compute. Expect faster, cheaper large-scale AI (e.g., high-accuracy OCR/PDF). Read More
Odyssey-2 (interactive video): responds as you type; streams new frames ~every 50 ms for real-time visuals. Read More
Cursor Composer (agentic coding): ~4× faster task completion with a new multi-agent workflow UI. Read More
Cartesia unveiled Sonic-3, a real-time conversation model using SSMs with just 190ms latency. It delivers expressive, natural speech in 42 languages. Read More
26th Oct – 1st Nov, 2025
Key Releases:
DeepSeek-OCR (open-source) compresses long PDFs into “vision tokens,” enabling high-accuracy OCR with ~10× compression, useful for digitizing handouts. Read More
OpenAI “Company knowledge” for ChatGPT pulls organization context from connected apps to answer with citations. Read More
ChatGPT Atlas (browser): Mac app puts ChatGPT in the browser with optional memory of visited content to improve future answers; available to free users. Read More
Copilot “Fall Release”: new Groups (up to 32 people), expressive Mico avatar, and “Real Talk” mode that challenges wrong assumptions, useful for collaborative classes. Read More
