Tell HN: Claude session limits getting small
I am a max subscriber for claude.ai using their browser UI and claude desktop ai app, not a API user, claude code user. In the last week I have come across every session of mine not lasting more than an hour or so. Have any of you come across this? Is this something new? Reaching out to their help is basically dealing with claude and canned responses, escalated to human but they've yet to explain/respond
16 comments
I have never had any of these problems using codex. Codex just works. on like the cheapest organization plan, I've never hit any limits, ever. Meanwhile I go on hackernews every day and I see you people talking about claude like its 1943 and you're frustrated that your cheese rations have been cut again.
claude has half the context window size of codex and blows through a good percentage of it right off the bat by injecting a system prompt the size of don quixote
I have been on Max + 20 plan for over 6 months without any issues that is why I’m wondering if this is something new.
Claude code sucks. Only noobs and X stans push it imo.
They mention specifically fixing this in 2.1.14:
https://github.com/anthropics/claude-code/blob/main/CHANGELO...
> Fixed a regression where the context window blocking limit was calculated too aggressively, blocking users at ~65% context usage instead of the intended ~98%
https://github.com/anthropics/claude-code/blob/main/CHANGELO...
> Fixed a regression where the context window blocking limit was calculated too aggressively, blocking users at ~65% context usage instead of the intended ~98%
i think it is happing way more quicker than ~65%, more like ~20% in my case. My sessions used to last me one week before, now 1 hour of plain back and forth stops the discussion without a single compact, no errors.
I stopped using Anthropic when they pulled a shitty fast one on Max users by limiting Opus usage and forcing people to use sonnet.
I’m happy with Codex 5.2 Extra high reasoning. It works imo way better, doesn’t bug out (CC has so many bugs), and I barely ever go below 40% weekly usage
I’m happy with Codex 5.2 Extra high reasoning. It works imo way better, doesn’t bug out (CC has so many bugs), and I barely ever go below 40% weekly usage
I have a few codes pro keys, in heavy refactoring on a large codebase I can blow through one account's weekly limit in a day. Its no longer about the limit though - a full day for $20ish is awesome - it is imo just better at everything than claude was.
Is it project-unrelated, that you ran into limits so quickly? If it's just in some repos, it might be a bloated claude.md. Looking forward reading from you, if you get some helpful insights
It is all related, it happens in the middle of a debug session where it just stops responding with no error message and my last message remains in the chat window.
i have to then start a new session and explain the context of pervious session and Claud goes on to ask for more context from the code and then by the time I get to any new code the session does what i explained above. i don’t allow Cluade to access my code directly its only via chat ui interaction. My usage data in the settings shows I have used less than 4% of alloted tokens
Ok. This isn't something I experienced, but it sounds strange. Are you using the Claude desktop client (electron app)? This thing is quite buggy and unreliable.. Or is the described behaviour in claude.ai browser input existent as well?
I’ve noticed the same thing, thought it was just me. Wonder if ralph wiggum would help.
Do you mean context size before compacting?
Yes, it never compacts anymore, just stops responding in the middle of a discussion. i thought it was one off but after 7 sessions of the same behavior it became very frustrating.
Just sounds like a bug. Try older version of CC.