Cursor is impossible to use at the moment

Started working at 7:00 CET, and Sonnet 4.5 is running fast for the first time in days.

Had 4 updates over the past 2 days.

I also discovered that a bug that was in an early version of Cursor has reappeared. The “Unable to reach model provider” error seems to happen very often in longer chats (although the context is not at all full). If I change to a new chat, it works. I remember this happened over a certain time in a past version (maybe a year ago?).

It definitely seems to be better, especially after upgrading to 2.1.19 on Linux, which arrived overnight. I remain surprised that Cursor didn’t treat this as a degraded service issue; their official status page for the last few days is squeaky clean.

@deanrie :

@deanrie : Hey deanrie, the error keeps persisting, but after testing it for a few days i think i have narrowed down the follwing:

Sonnet 4.5 = the error appears almost all the time
Gemini3 Pro = the Error appears after the agent runs sometime

MCP = If MCP is activated, the error appears way more often AND especially since i use the MCP Toolkit from Docker Desktop, with 50+ Tools in one MCP Connection the speed of the models is degregaded.

Same issues since Friday, I though it was temporary but no, I cannot work and I’m Ultra 200$/month…

It’s worst with Sonnet 4.5 which is the model I’m always working with. Same as others, I have messages indicating OpenAI provider problems, which must be one of the Cursor internal use of OpenAI which may fail, as I do not use GPT-5.1 when the failure happens.

Please @deanrie give some info about the failure and when you plan to release a patch or something. For the moment Cursor is not usable.


Request ID: 39789f71-2c4a-49e9-9d22-a57f511053c9

{“error”:“ERROR_OPENAI”,“details”:{“title”:“Unable to reach the model provider”,“detail”:“We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.”,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:false}

1 Like

Still happens to me too… honestly, since I do not have any feedback from the Cursor team and Composer has been working perfectly fine the whole time, I am beginning to think that the introduction of Composer introduced a conflict of interest with the Cursor developers. The company will benefit the most if users use their own models, so why make sure the other models work perfectly?

I am considering switching to the 200 $ plan at Anthropic and using Claude Code inside a 20 $ version of Cursor.

I second the above comment I’ve also been constantly having this issue and also upgraded to ultra thinking i would get better service but the errors are getting out of hand. Either messages dont go through and you get the ‘try again option’ or the message goes through and the agent is in the middle of the edit and the message pops up leaving the request incomple.. i dont think its fair that you’re charging us for a full request when the request has not been completed (and yes i have checked my dashboard and it does come out under usage) - i’ve also had several instances of this happening somes 2-3 times in a row which is insane. Its been about 6 days in a row now that every day im dealing with these errors, and If you even think to switch to another model performance isnt anwhere near the same as the context compression happens immediately and they often totally go off tangent so you end up having to use more requests just to get back on par.

Totally rethnking my subscription at this point, i was on basic and got forced into pro as the requests although supposedly the same would run out within 24 hours of refresh (no kidding, i have emailed several times about this and NEVER get a response), and even with pro the slowed service and overtime api rates feel exactly like the basic plan.. im now on ultra where im seeing some progress in requests but now whats stopping us is the actual model limits errors that are so frequent it’s basically like having a glorified pro plan.

From what i’ve seen this model limit error and the streaming cache error have been an ongoing issue since v 1.6

1 Like

I do not have any feedback from the Cursor team

A Cursor moderator responded at the top of this thread and shared it with our team. We’re aware, and can prioritize and quickly investigate reports that have complete instructions to reproduce. Thank you for those here adding those details.

I am beginning to think that the introduction of Composer introduced a conflict of interest with the Cursor developers. The company will benefit the most if users use their own models, so why make sure the other models work perfectly?

This is not accurate. We benefit when developers have a great experience with the product, regardless of the models used. Further, the top model used in Cursor this month has been Sonnet 4.5, so it would be bad for our business to not support these models well. Even further, our developers internally building the product choose and use from a variety of models, so we would not be able to do our job is we didn’t have access to many different models from different providers.

We’re actively investigating the reproduction reports here!

I really expect compensation for this issue because I haven’t been able to use Cursor for 5 days now. I contacted support 4 days ago and was told my ticket was “forwarded to a teammate” and that I should check the forum to see if others have the same problem. People clearly do.
I haven’t received any update from you since that email.
This means I’ve lost 5 days of my Ultra plan subscription and haven’t been able to use any of my credits throughout those days.

Please fix this and get back to me.

1 Like

Same issue with “Sonnet 4.5” model:

Request ID: 8aa2589c-2d37-4d2b-8a8d-5b6461a02e96
{“error”:“ERROR_CUSTOM_MESSAGE”,“details”:{“title”:“Model returned error”,“detail”:“The model returned an error. Try disabling MCP servers, or switch models.”,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}

I managed to get other agents working w/ the same query.

I’m back to having the same problems again this morning: {“error”:“ERROR_OPENAI”,“details”:{“title”:“Unable to reach the model provider”,“detail”:“We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.”,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:false}
ConnectError: [unavailable] Error . I have disabled my MCP, and the issue still occurs on Sonnet 4 and Sonnet 4.5 (this is with Cursor 2.1.26 / Linux x64). However, Opus 4.1 seems to work – this is not a sustainable solution given the model cost.

Recently, the OpenRouter models stopped working for me. At first, I thought the issue was only with OpenRouter, so I tried switching to the native OpenAI API with my own credits. I assumed that using the API directly would avoid problems, even though it would limit me to OpenAI models.

However, after adding credits to my OpenAI account, I discovered that they now require “higher-level internal reasoning,” which triggers mandatory Persona verification. I submitted all the required documents, but the verification is still pending. I’m based in India, so the verification process might also be slower for me.

I want to be clear that I’m not blaming Cursor for the Persona system, but it does feel like developers are being pushed toward using Cursor’s provided models and Cursor credits instead.

Right now, both OpenRouter and OpenAI are not accessible to me, which makes development difficult.

What doesn’t work:

  • https://openrouter.ai/api/v1 as base url (Seems intentional)
  • Direct OpenAI Key/Credits (due to internal higher level reasoning - not able to disable)

What else works:

  • Set “https://api.x.ai/v1” as base url (openai compatible) and use direct Grok credits.
  • Set Gemini direct API key.

I didn’t try yet:

  • Azure OpenAI
  • Claude Direct API

Hi leerob,

i appreciate you repling to my comment. It was a though that occurred after i did not hear from either the support or @deanrie after 3 days not being able to work with cursor at all.

Thanks for clarifying the standpoint of the developers and the company policy towards third party models.

I myself work with Claude 90% of the time, since it works best for me. I hope you understand my (an others) frustration over the current situation.

I am an early user of cursor and also a great fan, but on the other hand, are 200$ not a small thing for me as a freelance solo developer. And although i am all about early adopting and supporting companies i like, i still want a proper bang for my buck :slight_smile:

1 Like

same issue

Request ID: 3c5dd681-3d6d-41a8-a0b5-2f7ac514c558
ConnectError: [unknown] Network disconnected
at IQc.$streamAiConnect (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6330:405768)
at async vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:565:75254
at async nps.toolWrappedStream (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:2973:27159)
at async vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:4063:10091
at async KL.handleStreamComposer (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:2786:993)
at async dro.streamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6994:13536)
at async boo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7141:8433)
at async woo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7143:2281)
at async loo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7137:1113)
at async uoo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7137:33949)
at async foo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7137:48354)
at async X2e. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:2973:45096)
at async yoo. (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:7141:7461)
at async ose.processCodeBlocks (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:2899:3418)
at async Dj.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:4063:13780)
at async Lo (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:6207:3514)

Request ID: bdc4b970-c326-4bc4-a18d-d01407870243
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Unable to reach the model provider”,“detail”:“We’re having t

rouble connecting to the model provider. This might be temporary - please try again in a moment.”,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:false}
ConnectError: [unavailable] Error
at RZc.$endAiConnectTransportReportError (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6331:408452)
at KPo._doInvokeHandler (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6948:21873)
at KPo._invokeHandler (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6948:21615)
at KPo._receiveRequest (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6948:20377)
at KPo._receiveOneMessage (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6948:19194)
at yPt.value (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6948:17286)
at _e._deliver (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:49:2962)
at _e.fire (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:49:3283)
at Xmt.fire (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:6316:12156)
at MessagePort. (vscode-file://vscode-app/c:/Users/PC/AppData/Local/Programs/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:8973:18439)

1 Like