Your experience with subscription vs. token/API usage - Claude Pro "too good to be true"? (xpost from vibecoding)
retained text
Claude Pro is still an insane value. I'm worried it will soon be gone. It can't possibly be profitable for Anthropic. Last week I did a few tests and compared token-based / API inference vs. Claude Pro's weekly limits. If you max out your Claude Pro every week for a month (which is very easy) and you compare it to token-usage then (at least according to my own limited testing) the tokens are **20x-30x (!)** more expensive. I compared a few tasks in separate sessions, same model and reasoning, then looked at the weekly percentage used vs. the $ spent on tokens. Now you could say that the $20 subscription is just a gimmick because it's too restrictive. But I got used to it and I actually think the limits help me find a good, sustainable rhythm of planning, building, reviewing, and deliberating. It helps me get into a very productive flow. Compared to pure manual work, Claude Pro is already boosting my productivity by more than 10x. Maybe that's sufficient and even a reasonable ceiling, because it allows me to test, learn, and ponder, without the AI "leaving me behind".
document fetched 2026-10-22 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xbh2bg/your_experience_with_subscription_vs_tokenapi