Ask HN: What was the last task where only a frontier model could do it?

A Hacker News discussion asks users to share recent tasks where only frontier AI models (like Claude Opus or GPT-4) succeeded while open-source models (like Qwen or DeepSeek) failed. Respondents report that frontier models show meaningful advantages primarily in complex, sustained tasks like code disassembly and compiler work, though simpler reasoning and tool-use cases are largely comparable across model tiers.