Anthropic Just Made Claude Cowork Work With Any AI For FREE (+12 AI Updates)
This week brought major AI breakthroughs including China's free K2.6 model beating Claude and GPT on difficult benchmarks, OpenAI's launch of GPT 5.5 with autonomous planning capabilities, and ChatGPT Images 2.0 becoming the world's top image generation model. The video covers 12+ AI updates including new security concerns, DNA analysis capabilities, and various tool integrations.
Summary
The video opens with the presenter claiming this was the 'craziest week in AI all of 2026' (likely meant 2024), highlighting several major developments. China's Kimmy released their K2.6 model as a free, open-source AI that outperformed Claude and ChatGPT on humanity's hardest knowledge tests and coding benchmarks, while offering features like 12+ hour continuous coding sessions and coordinating 300 AI workers simultaneously. OpenAI officially launched GPT 5.5, described as their smartest and most intuitive model, with improved autonomous planning capabilities that allow it to handle complex multi-step tasks without constant user guidance. The presenter discusses a significant security breach at Versel (valued at $9.3 billion), where hackers gained access through a small AI productivity tool installed by an employee, stealing API keys and employee data worth $2 million on the black market. Other notable updates include a person using AI to decode their own DNA at home for $1,100 in 4 hours (compared to the previous $3 billion, 13-year hospital process), SpaceX's collaboration with Cursor AI involving massive compute resources and a potential $60 billion acquisition, and Elon's new Grok 4.3 that can build PowerPoint decks and analyze videos. The video concludes with an extensive demonstration of ChatGPT Images 2.0's capabilities, including brand continuity across multiple images, working QR code generation, multilingual text rendering in Japanese manga, and advanced image editing features that allow frame extraction and recomposition.
Key Insights
- China's K2.6 model leads every major AI model including Claude and Gemini on humanity's last exam, considered one of the hardest AI knowledge tests, while being completely free and open-source
- The Versel security breach occurred when hackers attacked a tiny AI productivity tool that one employee had installed, bypassing multifactor authentication and stealing data on 580 employees worth $2 million
- SpaceX's deal with Cursor AI involves access to roughly a million H100 equivalent GPUs and gives SpaceX the right to acquire Cursor for up to $60 billion in 2026
- A person successfully decoded his own DNA at home using AI for just $1,100 in 4 hours, compared to the previous cost of $3 billion and 13 years in hospitals
- ChatGPT Images 2.0 can now maintain character and object continuity across multiple images, create working QR codes, render accurate multilingual text, and perform advanced image editing including frame extraction and recomposition
Topics
Full transcript available for MurmurCast members
Sign Up to Access