OpenAI is lying

A developer rants about OpenAI's blog post claiming GPT-4.5 can produce production-ready frontends with the right prompting. The author argues this is misleading, showing benchmark comparisons where GPT models consistently produce repetitive card-heavy layouts regardless of design skills or prompts. Competing models like Claude Opus, Gemini, and Kimi K2.5 are shown to produce significantly more varied and usable UI outputs. The author speculates the root cause is insufficient or outdated training data for UI tasks in OpenAI's models, and criticizes OpenAI for publishing a misleading article that erodes developer trust in AI tools.