Daily Highlights
1. NVIDIA's Open Physical AI Ecosystem Expands: NVIDIA made multiple announcements at GTC Taipei, significantly bolstering its physical AI offerings. The company launched NVIDIA Cosmos 3, an open world foundation model for physical AI, which is now available on Hugging Face and GitHub. This "omnimodel" integrates vision reasoning, world generation, and action prediction. Complementing this, the NVIDIA Agent Toolkit was unveiled, providing open-source software and blueprints for building secure AI agents, featuring the Nvidia NemoClaw framework and the OpenShell Secure Runtime. Additionally, NVIDIA Alpamayo 2 Super, a 32-billion-parameter vision language action (VLA) model for safe, level 4 robotaxi development, was introduced, alongside new tools like AlpaGym and OmniDreams.
2. MiniMax Debuts Multimodal M3 with 1M Context: MiniMax released its M3 flagship model, a significant entry into the AI landscape. This model is touted as the first domestic AI to combine frontier coding, agentic capabilities, a 1-million-token context window, and native multimodal processing within a single architecture. Benchmarks indicate "industry-leading" performance in coding and agent tasks.
3. Google Gemini Flash Deprecation & GitHub Copilot Pricing Shift: Developers using Google's AI platform faced a mandatory migration as Google retired gemini-2.0-flash-001 and gemini-2.0-flash-lite-001 APIs, requiring an immediate update to gemini-3.5-flash. Concurrently, GitHub Copilot transitioned to an "AI Credits" billing model, where most interactions, including chat and autonomous tasks, now consume metered tokens, potentially increasing costs for heavy users, though core code completions remain free.
Niche Project of the Day
The NVIDIA Cosmos Coalition was launched today, a global collaboration between world model builders and AI developers, including founding members like Runway. This initiative aims to advance the next generation of open world models for physical AI through shared infrastructure and mutual technical contributions, with the first project being a base model co-developed by Runway and NVIDIA.