Just took a look at @TencentAI_News latest update, and the pace of progress is seriously impressive. Three major releases in 6 months, roughly one every two months. Hy4 preview is already pushing into the top tier of open-source models. But the more interesting part isn’t just the parameter count. It’s Co-Design approach: use a widely used agent like WorkBuddy to collect real-world feedback, feed that back into rapid preview iterations, then use the stronger model to improve the product experience. Its internal blind tests reportedly edged out GLM 5.3 and Kimi K3 across 200+ tasks. That model + agent feedback loop is starting to look like a real flywheel. I must say “Chinese Codex” positioning is becoming increasingly hard to ignore.