Yesterday, I used DeepSeek v4 Pro intensively to mess around with some agentic tasks. Checked the bill today—only spent 2 RMB, with a cache hit rate of 92%+. This price is absolutely insane. If this were done on Claude Sonnet, the bill would be at least an order of magnitude higher. In the past, DeFi brought financial equality, and now open-source large models are bringing token equality. When inference costs drop to just a few cents per round, it benefits millions of independent developers, students, and small teams around the world who want to do something with AI but have limited budgets. When tokens are cheap enough, regular developers can afford to let agents run a few more rounds, break down a few more subtasks for parallel processing, and not have to meticulously calculate every step to save tokens.