Anthropic/OpenAI may be spending more than $1000 for every $100 you pay them

Anthropic and OpenAI's subscription models heavily subsidize the true costs of using large language models for intensive tasks like coding—users paying $100 monthly for Claude Max could consume tokens worth over $1000 at actual API pricing if they fully utilize agentic coding features. The author argues that while LLM-assisted coding is impressively capable, it is not economically viable at scale without these subsidies, and the companies' efforts to reduce costs signal diminishing returns rather than continued improvement. The fundamental problem is that serious applications requiring complex reasoning and tool-use remain prohibitively expensive compared to what consumers are charged.