At 00:30 a.m. on June 3rd, Microsoft took the wraps off its first self-developed cutting-edge inferential AI model, MAI-Thinking-1, at the Build 2026 conference. This model is no slouch, packing a whopping 35 billion active parameters. What sets it apart is its training approach—it's designed without relying on knowledge distillation and is trained entirely from the ground up using pristine, untainted data. In the same vein, Microsoft also rolled out MAI-Image-2.5 and its Flash version image generation models, which are top-notch when it comes to crafting and tweaking images. Moreover, Copilot is set for a major makeover, bringing together a host of Microsoft AI assistants under one roof. There's even talk of incorporating the Scout AI agent, though we can expect the beta version to hit the scene by late summer.