Unable to Use Claude 3.5, 3.7, and GPT-4o Properly in Cursor – Code Too Long / Connection Failed

[gaokui2010]

Hi,

I’m currently using Cursor and facing issues when editing large code files (around 780 lines) using Claude 3.5 and Claude 3.7 models. When I attempt to modify code longer than 600 lines, the response is cut off early with the message:

“Model provider Anthropic ended response early.”

The issue persists after many attempts. Here are the Copy Request IDs for reference:

Claude 3.5: 7084bec0-11cb-428a-a6f8-38dfce828565

Claude 3.7: 3540bbc2-ba9c-4799-9fe2-329cc3431749

I also tried using GPT-4o, but it returns:

“Connection failed. If the problem persists, please check your internet connection or VPN.”

I followed the troubleshooting guide and disabled HTTP/2, but the issue still occurs. Currently, none of the models work properly from the Toggle menu.

I would like to mention that I am not using a VPN. My network is in China. My operating system is Windows, and I am using the latest version of Cursor (I subscribed to the Pro service via the official website just yesterday). Since subscribing to the Pro service, I have been unable to use Cursor’s AI services effectively. I have been facing these issues since I began my Pro subscription.

Could you please help resolve this issue? I’m unable to work efficiently because of these limitations.

Thanks in advance!

[ahadda5]

Facing the same issue. I think it started with v.49
There is definitely something about the context and Thinking model. (Ask feature does not seem to have that issue).

First error message always starts with this.

Then one can get the connection failed error.

I cannot pinpoint the issue.

Finally, the requests are showing up in my cursor profile , so no usage limit is reached.

Version: 0.49.6
VSCode Version: 1.96.2
Commit: 0781e811de386a0c5bcb07ceb259df8ff8246a50
Date: 2025-04-25T05:07:16.071Z
Electron: 34.3.4
Chromium: 132.0.6834.210
Node.js: 20.18.3
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.3.0

[zyc-bit]

The same.
I am in China too, facing the same problem (and I subscribed to the Pro Service via the official website today)
Anyone has solution?

[ahadda5]

Same issue

[ahadda5]

Any updates! i seem to hit that limit around the 3rd week of everymonth. My profile shows i still have lots of requests remaining even for premium models.