Yeah this is a bit crazy and not surprising at all.
The limits have always been opaque and you never know when they change.
I started building an open-source local proxy that logs every rate-limit header Claude Code sends.
I am using it to track and get a better sense of the 5h and 7d weekly limits.
Some initial data from 11 observed 5h sessions on Max 20x:
- 5h budget: roughly $120–$280 per window
- 7d budget: roughly $1,300–$1,900
- Separate Sonnet-only 7d budget at ~$150
- 95% of tokens are cache reads. They barely move the meter.
It’s open source so more people can run it and we can figure out the real numbers.
This morning I hit 100% 5hr usage on a task that took ~10% in the past. Looks like they are still testing the limits, but it seems over-tuned to me.
Also not great that they communicate this now, since people have been complaining about sudden and strange usage spikes for a few days with no response from Anthropic.
One of the shadiest things I've seen is how openai treats their weekly limit. They reset it whenever they want! So if you use 25% but then it's day 3 or day 4 of the week, you've used less than the pro-rata, they'll just reset the limit.
Makes me incandescent that OpenAI would have these different rate limits that apply to you, but they can do whatever they want with your limit. It's just incredibly hostile treatment & incredibly incredibly rude.
I'm on a plus plan, not max. I probably would go through around 3 maybe 4 plus plans a week; I use glm-5 pro plan (from before they added weekly limits) a lot more.
I’m on the Plus plan too and never run into limit issues. It’s one of the main reasons I stay subscribed because I feel I get my moneys worth. I love Claude models but usually feel cheated after not using them (especially Opus) for very long before I’ve hit a limit and they’re bilking me for more.
It's crazy to me that this is not considered fraud. You sign up for a yearly plan under a given assumption of functionality, then they just change the terms to give you less than what they agreed to without compensating you in any way. That's textbook fraud.
Like electric cars. It's nice to charge it in summer 2013 when you're the only one on supercharger, and pain waiting in a pissing queue of 10 cars right now
I keep saying it because it’s true: I do an insane amount of work with my little $20/mo. ChatGPT Plus subscription and never hit limits. For me Claude (especially Opus) is not built for real work, no matter how good the model may be, because the limits are comically prohibitive. Which is a shame because I love their models, but their shadiness around usage is bad business.
The limits have always been opaque and you never know when they change.
I started building an open-source local proxy that logs every rate-limit header Claude Code sends.
I am using it to track and get a better sense of the 5h and 7d weekly limits.
Some initial data from 11 observed 5h sessions on Max 20x: - 5h budget: roughly $120–$280 per window - 7d budget: roughly $1,300–$1,900 - Separate Sonnet-only 7d budget at ~$150 - 95% of tokens are cache reads. They barely move the meter.
It’s open source so more people can run it and we can figure out the real numbers.
https://github.com/abhishekray07/claude-meter