Why are Claude 3.5/3.7 slow requests so extra-slow now?

My current recommendation is to use a different agent to map everything out into a copy-able MD prompt for planning and what not, then paste that prompt into claude and do the waiting game. There are other similar services to cursor, but I think Cursor is the best for now. As long as they fix the slow requests, lol.

1 Like

Mate, coming with some bad news.
I made some research, and found that such behavior is intentional. Yes, thats misleading users, but that’s what they do. Reddit user made his research, and found out that the “queue” is artificial. There is no queue, but there is artificial slowing down of requests, that has been confirmed on Cursor forum (1st link).

I will slowly switch to other tools, copilot in VS code seems to be working OK with Claude.
There is no point in trying to get answer from devs or mods, they will just say how sorry they are etc. Best solution is just to use what they want you to use - usage based pricing, or to switch to another tools.

You can read more on:

After exceeding 500 requests, what are the rules for slowing down? 600, 700, 800, etc

Cursor needs you to continue recharging.

Alright fellas. Official response. Time to burn it down lol. Seriously though will probably go ahead and cancel to go with windsurf or roo code or something else. Anywho, no need for me to remain here at this point. You all take care.

1 Like

Yes, I really didn’t expect that the slow requests from pro users could be so slow. cursor really surprised me. It’s better to run away earlier. You can use AI studio or trae

2 Likes

M gone give it a try I paid for a lot of premiums 3.7 thinking request and its is performing poorly and keep deviating from context. I tired directly and it was keeping context. seem like in cursor something is intercepting the context and causing issues even iwth large context enabled.
I have cancelled my subscription for now until this changes.

2 Likes

Or just create as much gmails as you need and use cursor :slight_smile:

Oh wow… I was about to complain about Claude 3.7 being so extremely slow as a Pro user (several minutes per reply), but now that I’ve read this I realize why.

It was possible to compensate for this for a while with Gemini 2.5, but for the last 1-2 days it has been slower and slower and above all “dumber”, “lazier” and regularly does not make any changes, although it says it does.

In short: Cursor has become practically unusable for me. And when I read here that response times are deliberately slowed down, then it’s time for me to look around at the competition. This is ZERO fun.

Hey, just to clarify here, nothing has changed around how this works!

Once you have exceeded your fast requests, your next request will only have a wait of a few seconds. As you use more and more of these slow requests, they will incrementally get slower.

By the time you hit 2000 slow requests (4x the usual allowance), it would not be unexpected to see >60s wait times, as this is a huge increase in volume vs what you pay for.

The slow requests pool was always intended as a backup so that users didn’t have to worry about using all their fast requests, and then being left without premium model access for the rest of the month.

The reason why the delay is somewhat artificial is that actually creating a working “queue” for thousands of requests each minute would likely be more of a negative that a benefit - therefore, we employ this system to ensure we safeguard our capacity for fast requests!

If memory serves me, the delay in slow pool requests is somewhat model dependant, so using models other than Claude may help you have a shorter wait time!

Feel free to shoot any questions here in the replies.

3 Likes

I mean, I just don’t get it, how come all of a sudden the requests started taking two minutes, when literally just a moment before they were kicking off in a matter of seconds? If things are really the way you say they are, then what could’ve possibly changed to make them slow down like that? I was working too, and between one request and the next claude literally went from three seconds to two and a half minutes — with no in-between, just all of a sudden out of nowhere. And I was already working in slow mode.

yes i see that too last 2 days

Hey! Just circling back on this, totally understand if you’re busy, but I’d really appreciate any insight when you get a chance. Still trying to figure out what might have caused that sudden jump in delay times and I hope the team is working on the issue. Thanks in advance!

The real move is to subscribe to both cursor and windfall…get fast mode from both…and slog it out for the remaining few days in the month.

The pricing models are a cash grab, now that there is some serious competition, cursor going to have to get real with its prices

1 Like

Basically, you hit the limit that they have set for heavy users, so you’ve been throttled.

Basically, applying a hard limit as a 60s wait is not usable by any mean. The way Cursor deals with customers and capacities issues is terrible. 0 communication, continuous degradation of the service for paid customers. You may have priced your service wrongly, that s not a reason to make those changes without any proper communication

93e19330-f1d4-4d51-93d3-dc953c483486
Could you please check if this is correct?
So slow