Claude Code users hitting usage limits 'way faster than expected' | Hacker News Reader

Claude Code users hitting usage limits 'way faster than expected'

bbc.com

22 pointssteveharing118 comments

I got curious where my tokens were actually going, so I wrote a quick audit tool that reads the Claude Code session logs.

Turns out the biggest sink isn't your prompts. It's the agent's own tool calls. In my case, grep alone ate 3.5M tokens across ~350 sessions. 1800+ calls, most of them dumping raw output that the agent barely used. Full file reads for one function signature. Complete test output when only failures matter.

So I built wrappers: function signatures without bodies (~90% smaller), test output with just failures, that kind of thing. 2.3M tokens saved so far.

You can audit your own sessions without installing anything:

  npx tokenlean audit --all --savings

https://github.com/edimuj/tokenlean

Neat! Installed this for Claude Code and am adding Codex support to offer back, assuming it works out. Great tool idea and looking forward to seeing it in operation over the next few days. The 'audit' command is a great feature idea more tools should integrate.

can I use it with rtk and headroom?

For a start they could make the answers less talkative?

I switched back to ChatGPT out of necessity, because Claude stopped working after two queries, where it gave overly elaborate answers (about a simple web app config).

But Claude isn't alone. It seems a recent (subjective) trend that Claude and ChatGPT give very lengthy answers, with a lot of repetition from the original query on the free plans.

I got used to add "answer briefly", to keep the noise in check.

And just as with a real human rambler, the longer they rambler, the more likely they are to start making stuff up and asserting false truths

Yes lately i've also noticed the same pattern that Model try to provide over explaination to even simple stuff & that points to its system prompt or something internal instructions to waste tokens to hit limits

> Anthropic recently accidentally released part of its internal source code for Claude Code due to "human error".

I wonder who that human was counting on leading up to this "human error" ...

Yeah the whole OpenAI exodus brought in a ton of people and Anthropic was struggling to meet the previous usage already

That’s why there’re now work hours restrictions

Yes that make sense also Since Anthropic says other Chinese companies using their data for their models, they might be limiting use on new accounts.

How ironic: once the exfiltrators of all of the web's data have consolidated it into their own walled-garden it becomes 'proprietary' and must - of course - be protected from exfiltration by others as if it was their own.

This is something these tech giants ignore intentionally. Infact many people don't even know about how they train their model by scraping data for free & when it comes to their code being open source, you see Takedowns lol. Interestingly, Anthropic made a bigger Contribution to open source itself.

Not their data.

Is that really on BBC? what a world we live in...

Anthropic launched in the UK recently (Feb I think) so I expect it’s as a consequence of that.

Yes it could be possible

There are ways to reduce token usage if you use Claude correctly :)

Reply on news.ycombinator.com