What do you actually get when you pay for Claude Code? That question keeps coming up, and Anthropic has not made the answer easy to find.
The latest round of confusion centers on whether Claude Code has any kind of unlimited usage tier, or whether developers are misreading how token consumption works. Unlike ChatGPT Pro's $200/month plan, which OpenAI markets as essentially uncapped, every Claude plan has usage limits - even the $200/month Max 20x tier.
Here is how it actually works: Claude Code uses a 5-hour rolling window system. Pro subscribers ($20/month) get roughly 44,000 tokens per window. Max 5x ($100/month) gets about 88,000 tokens. Max 20x ($200/month) gets approximately 220,000 tokens. Once you hit your cap, you can continue by paying standard API rates, which is a separate billing system from your subscription.
The problem is not the limits themselves - it is that Anthropic does not show per-prompt or per-token breakdowns in the interface. Developers regularly report hitting caps mid-session without warning, unsure how much of their quota a particular task consumed. When you are deep in a coding session and the tool suddenly throttles, the instinct is to wonder if something is broken rather than assume you burned through your allocation.
Anthropic added weekly quotas in August 2025 on top of the rolling window, creating another layer that heavy users run into. The interaction between the 5-hour window and the weekly cap is not well documented, which feeds the cycle of confusion.
For anyone using Claude Code seriously, the practical advice is simple: if you are on a Pro or Max plan and doing sustained coding work, expect to hit limits. Budget for API overflow costs or plan your heavy sessions around the 5-hour reset window.