Summary
This weekly AI news roundup covers a high-density slate of releases spanning model development, image generation, video tools, robotics, and platform security. Microsoft releases Mage Flow, a 4-billion-parameter open-source image generation and editing model with built-in ControlNet, depth mapping, segmentation, and a turbo variant that generates images in under one second. ShotPlan enables multi-shot video generation with precisely timed cuts, fades, and camera movements specified through text prompts.
On the model front, Anthropic releases Claude Opus 5, positioning it as a near-frontier everyday model for knowledge workers and enterprises at roughly half the cost of comparable models. Google releases updated Gemini models with improved speed. Alibaba teases Qwen 3.8, a large upcoming open-source frontier model, and releases Qwen Image 3 as their strongest image generation model to date. The text-only GLM 5.2 gains vision capabilities. A notable security incident sees HuggingFace targeted via OpenAI’s internal models, with the open-source GLM 5.2 reportedly helping detect and remediate the issue.
Robotics coverage includes Unitree’s Super Athlete AS2W, a wheeled-legged robot that reaches 6 m/s, carries 16 kg of payload, and operates in temperatures from −20°C to 55°C. The episode also covers a new open-source interactive world model for Minecraft. The video is sponsored by Higsfield, an all-in-one AI content creation platform supporting models including Seed Dance, Cling, and GPT Image.
📺 Source: AI Search · Published July 26, 2026
🏷️ Format: Roundup
