OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

The thread treats the cuts less as a mysterious act of generosity than as a collision between competition and rapidly improving inference economics. Several commenters point to Chinese and open-weight alternatives, especially options that are cheaper and less prone to refusing cybersecurity-adjacent work, as pressure on OpenAI and Anthropic. Others add that better hardware, separate training and inference chips, KV-cache improvements, smaller deployed models, and higher serving efficiency can create real room to lower prices without necessarily selling every token at a loss. One participant cites throughput-and-revenue estimates from a competing provider as evidence that profitable inference is at least plausible. The larger argument is about whether cheaper tokens signal healthy commoditization or a market headed for a wreck. Optimists expect LLM access to resemble cheap, abundant connectivity: lower prices unlock automated agents, hobby projects, and broader adoption, while providers recover costs through volume and efficient infrastructure. Skeptics say token generation retains an unavoidable energy and hardware cost, so unlimited plans and repeated price cuts cannot work like mobile data. They distinguish belief in useful LLMs from belief that today’s frontier labs have viable economics, warning that price competition may destroy vendors before it benefits users. A secondary concern is that regulation could be used to exclude Chinese or open models rather than improve AI safety.