This AI news roundup highlights NVIDIA's launch of the open-source Nemotron 3 Ultra, a 550B MoE model optimized for long-running agents, and Anthropic's internal data showing Claude now authors over 80% of merged code, indicating early signs of recursive self-improvement. Cloudflare acquired VoidZero to strengthen its agent-friendly developer platform, while OpenAI's ChatGPT surpassed 1 billion monthly active users. The update also covers new agent evaluation infrastructure, open image models like Ideogram 4.0, and frontier AI adoption signals including a joint letter on biosecurity screening.
Microsoft at Build 2026 announced seven new MAI models, including the flagship MAI-Thinking-1 reasoning model with 35B active parameters, 256K context, and strong benchmark scores like 97% on AIME 2025. The company released a highly transparent 109-page technical report that impressed researchers, emphasizing clean data lineage and no use of synthetic data or distillation. Build also focused on local AI with Windows as an agent runtime, the RTX Spark Dev Box, and Project Solara/Scout agent hardware. The GitHub Copilot app was unveiled as a desktop home for agent-native development, and Web IQ was introduced as a new grounding API for agents. Overall, the event positioned Microsoft as both a first-party frontier model developer and a multi-tier AI platform company.
Ethan He argues that video models' intelligence primarily comes from LLMs, not video data, and that video agents are the next major evolution in generative media. He describes building Grok Imagine from scratch in three months at xAI, emphasizing iteration speed and debugging data pipelines over new algorithms. The conversation covers the high cost of storing and moving video data, step distillation for fast inference, and challenges in audio-video alignment. He predicts that video agents will reach production-grade quality by the end of the year, surpassing standalone video models.