[Rumor: OpenAI's Internal Testing of GPT-5.6 Achieves Breakthrough in UI De-Slop, Default Output Generates Minimalist Lumen Notes] According to monitoring by Beating, leaked internal testing screenshots shared by blogger leo reveal that OpenAI's next-generation large model, GPT-5.6 (internally codenamed iris-alpha), has made significant progress in front-end interface generation, achieving a notable breakthrough in UI de-slop. In its default state, without any UI guidance prompts, the new model directly generated a minimalist and meticulously designed interface for a note-taking application called Lumen Notes. Compared to the cluttered and formulaic web pages typically produced by generative AI in the past, GPT-5.6 demonstrates an exceptionally refined sense of grid layout and design aesthetics. It shows remarkable restraint in spacing control, font weight hierarchy, and the subtle use of lavender tones for key components. Just ten days ago, leo had publicly criticized GPT-5.6's poor front-end generation capabilities, describing the default output as extremely chaotic in terms of layout. Yet, only a week later, the new version screenshots shared by leo showcase a revolutionary de-slop effect, proving that OpenAI is actively addressing the shortcomings in front-end code generation. The new model is expected to be officially released in June, at which point it will directly compete with Anthropic's Claude in delivering Artifact-based interactive generation experiences, offering users high-aesthetic, reusable front-end code. [Original Link]