Claude 4.5 Sonnet <thinking> no longer thinking?

Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

About…maybe, 40 minutes ago now, it appears that Claude 4.5 Sonnet Thinking, simply stopped thinking. I am still on the thinking version of the model, but it is just no longer thinking. Its non-thinking self seems utterly incapable of handling any of the instructions I am giving it, and it cannot seem to solve even rudimentary problems right now.

I don’t know when, exactly, thinking stopped. It was indeed working earlier in the chat, with numerous 1s, 2s, and occasional 3-5s thinking cycles. Then suddenly, it would think once at the beginning of responding to a prompt, then stop thinking until it had completed the work. Then it reached a point where it wouldn’t even do a single thinking cycle.

It seems as though all reasoning ability of this model has vanished. However, it does not seem to be performing even at the same level as claude-4.0-sonnet non-thinking…without its reasoning cycles, it can’t seem to perform at all.

This is kind of the cherry on top, of a now months-long cycle, of continuous and progressive degredation in Cursor for me. Back in 1.3, aside from the terminal issues, the rest of Cursor was working wonderfully. @Docs, my favorite feature, worked superbly. Could use @Web, @Git, past chats, etc. Over the last two months or so, one by one, these features have progressively broken. Even recent features, such as sonnet 4.5, and the new custom slash commands feature, are now breaking…and they have only been around for a few weeks? Maybe just a week?

I’m totally baffled, extremely frustrated, and honestly starting to wonder what my $200+ a month is buying me, if cursor features are going to continually degrade, fail, and break completely. I rely heavily on @Docs, and it still seems to be broken. I really need the ability to reference past chats, for certain pieces of critical work, and cannot. I recently delved head on deep into custom commands, and those suddenly stopped working. Now sonnet 4.5…seems to have degraded into RetardLLM.

Is there any way, I could actually TALK with someone at Cursor, to help me resolve this progressive and ongoing issue? I just chatted with a coworker, who seems to be on the latest version of Cursor, and things I’m having problems with, work for him. I am honestly now starting to wonder, if perhaps there is an account-level issue that is a part of the problem here? If my instance of Cursor degrades any more, it will cease to be a useful tool and I have come to rely on its features very heavily over the last 6 months, and I really need to get the value out of what I am paying for…

Steps to Reproduce

Don’t know…not sure what caused the problem, it just simply seemed to happen progressively over a period of 30-45 minutes…

Expected Behavior

Thinking models should think…

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.7.52
VSCode Version: 1.99.3
Commit: 9675251a06b1314d50ff34b0cbe5109b78f848c0
Date: 2025-10-17T01:41:03.967Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.5.0

For AI issues: which model did you use?

Sonnet 4.5 Thinking

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

2

Hi! I’m seeing the same thing. After today’s update, the Claude 4.5 model and Auto mode stopped thinking altogether. The results dropped a lot. I tried GPT-5 and it still thinks, but Auto and Claude 4.5 suddenly got much worse.

3

Hey, thanks for the report. I can confirm multiple users are affected, several reported the same issue after today’s update, and there have been multiple reports of Claude 4.5 performance degradation in recent weeks.

Could you share a request ID? To start, you’ll need to turn off Privacy Mode.

4

d0f5f1d7-7946-4c0d-81a4-587c0b368a76

1 Like
6

It looks like Cursor updated this morning, and I am now on 1.7..53. ATM, claude seems to be working, and does appear to be thinking. If it stops again, I’ll try to grab a request id.

1 Like
7

FWIW, it also seems as though the performance of the claude models has absolutely tanked. When Sonnet 4.5 first hit, it was much faster than its predecessors. It was not as fast as Grok Code, but much faster than it used to be.

It now seems to be as slow as Sonnet 4 was, and at times it seems to be much, much slower. This slowness has been ongoing since late last week at least, but the last couple days it does seem much worse. Not sure if this is attributed to the AWS outage or not, however today, Claude, Sonnet 4.5 in particular, seems to be much slower than it was before.

Is this an issue with Claude, or is it an issue with Cursor?

Also, FWIW, it appears the model has once again stopped thinking. I have privacy on at all times, so, that kind of makes it hard to capture request ids with privacy turned off, as we need privacy here. So I am not sure what to do about this. In any case, model performance, in multiple ways, has tanked to a new low today.

1 Like