Anyone else hitting the limits earlier than expected?
I’m on the Pro Plan. This morning I ran two queries. The first one was a daily briefing and the next one was to revise that skill slash index and I came up against the limit with those two queries. I have to wait until 11am to continue.
That seems a bit premature to me.
Anyone else having issues?
Using Opus?
In settings there’s a Usage tab
I use Claude regularly all day, and have very rarely blown through limits. For the cases where I need extra fuel I have “Extra usage” turned on. My credit card is configured, but I keep auto-reload offer. I think Claude has asked me once in six months to top off the extra usage tank.
Katie
Ugh. I have it set to an additional $20 - it blew through that in about an hour.
Yes. Opus 4.6 but regardless of what I use, I think it’s a bit crazy I’m blowing through it so quickly.
Opus is like a large V8 gas guzzler ford truck
Have you asked Claude to evaluate the high-cost prompt, tell you the reasons it costs what it costs in tokens, and have it suggest revisions to the prompt(s)?
Katie
You probably shouldn’t use Opus much on the $20 plan. On the $100 plan, you can stay on Opus almost all of the time. Sonnet is fine for most tasks.
I am on opus all day, haven’t noticed a change as of late. Possibly depends on what you are doing, has the nature of your work changed?
It was all my fault😉
Free is hitting buffers quickly now for me.
I’ve gotten in the habit of asking Claude if Sonnet is up to the task at hand or if I need Opus. Most of the time, it recommends Sonnet.
I thought, “This is not real. I’ve never hit my limits.” But of course, yesterday I hit my limits with Claude Cowork without doing a whole lot; I was using Sonnet. I guess Anthropic has been playing a little loose with limits.
"To manage growing demand for Claude we’re adjusting our 5 hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged.
During weekdays between 5am–11am PT / 1pm–7pm GMT, you’ll move through your 5-hour session limits faster than before."
Yes, all the time. It’s somewhat annoying when you’re paying $20 a month.
I bumped into the limit today using Sonnet 4.6. Donkey work data entry for waste disposal records. I have Cowork reading and renaming files, comparing the handwritten data via OCR to a summary file, then taking all of the data and populating a spreadsheet for the annual waste summary.
It has runs some conversions, and today it saw that I did not have a tab set on the facility file for 2026, It read previous year, and then the most recent file that it had worked on, set and populated the 2026 tab noting a duplicated column. Time cost burning through my pro plan plus a $1.50. I don’t know if it is the OCR review that is eating time, but each site is running smoother and $21.50 is cheap compared to the opportunity cost of my admin entering the data. She will get moved to QA\QC work on the final product.
The longest part of the cycle is the manual downloading of files from a government website - I can only download one file at a time. (Claude is only allowed to work on files in very specific folders on my machine).
I asked Claude and this was an interesting answer:
“What eats your quota fastest:
Your usage is affected by the length and complexity of your conversations, the features you use, and which Claude model you’re chatting with. Long threads are a big culprit — Claude takes in all the history of a conversation as context, so large context length greatly reduces your usage limit. Starting a new conversation can help you avoid reaching the limit too quickly.”
I’ve been making the mistake of just using one long thread! Might be a solution to use new queries for new topics.
Definitely. Now that 1MM context is generally available, you can accumulate more in context than previously, too.
Longer thinking, agent spawning, and larger context carrying are probably responsible for most people’s increased usage, and if you don’t catch those things, it feels like you’re allowed less usage.
Glad I found this thread. Claude has just locked me out until 1st April, 8pm. I’ve been busy setting up my ‘robot assistant’ (I bought David’s excellent field guide on it and have been busy switching over). Now, I’ve been doing the setup tbf so probably burnt through a lot of tokens, but I’ve also seen a lot of disgruntled users on reddit who are bumping up against limits.
I wonder if @MacSparky might be persuaded to consider adding a ‘how do I not burn through my tokens with the robot assistant’ workshop to his planned workshop series as part of the course? I’m only on the pro plan, but I imagine quite a lot of us are.
That said, if anyone would be kind enough to recommend a decent guide to making your Claude use as efficient as possible, I’d very much appreciate that.
I was too dismissive; there is some kind of bug/misconfig that they’re looking into this afternoon. (Tweet is by an Anthropic employee.)
Yes, this is happening to me too!