ALL models redirect to fake 3.5 Sonnet

[peachyGT]

No matter which model you select or type inside the box, the script automatically redirects the model to a different one. The model says it's 3.5 Sonnet when asked, however it seems to be the cursor-mini model instead, or a similar small sized model (could also possibly be GPT 3.5 turbo), as the response quality is significantly diminished from regular 3.5 sonnet.

This issue was referenced in #50 #47 #46 , but never adressed directly. I encourage the use of this thread as a centralized discussion point for this specific issue.

Edit: Should be noted that the issue does not occur in cursor itself and has not been reported on cursor's official forums. The issue is tested both on Risu AI and on SillyTavern - it is consistent in both frontends.

[jaimesebsf]

Confirming what you say, I asked him and at first he only mentioned inconsistencies and if you mentioned Claude, Gemini or any model he told you it was one of them and trying to ask without mentioning any AI model just asking for his, he answered this.

[oadwpadonwa]

Confirming what you say, same #

[JiuZ-Chn]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5.
I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

[hjutahai1212]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

I use 2 proton Emails and one Gmail. I had one proton email where it worked normally for most of the time. But since like ~4 days it started randomly giving me my preferred model or the redirect. The other proton mail and Gmail were mostly on and off redirects. But they all were working in the Cursor app while they weren't in SillyTavern.

Edit: I don't think it's flagging us as abnormal accounts since it's working in the Cursor App

[jaimesebsf]

I use 3 Google emails and this redirection has been in place for about 3 or 4 days.

I change from the models I choose (Claude 3.7 or 3.5) to this model that says inconsistencies

[oadwpadonwa]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

1. I'm using gmail accounts

2. The same issue

[peachyGT]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

1. I use exclusively gmail accounts via cursor, and had not run into any issues until 2-3 days ago when the model started downgrading.
2. Cursor app itself allow use of 3.7, but using the script consistently causes this issue.

[Napalm2]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

1. I've always used Gmail accounts, all created a long time ago

2. Same problem

[Trastor22]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

1. Using 3 Gmail Acocounts I have since 10+ years ago

2. Yup same problem

3. Siento que arranco la carretera, Voy enfierrado por la costera, Mi troque casi se desparpaja, Pero pura madre se me raja

[udjhhdk828373]

This is a good catch, please share:

1. What kind of email do you use (gamil? outlook? temp mail?)
2. Does the account behave the same when using the script and Cursor App? For example, if the script fails to work with 3.7, can Cursor App still work with 3.7?

The model confusion issue of this model is not a new issue. Cursor used this strategy to "ban" abnormal accounts in February, forcing abnormal accounts to use GPT4 and using prompts to make them think they are Sonnet 3.5. I think this may be caused by the script's disguise strategy not being enough to fool Cursor App, but more research is still needed.

1. Using 3 Gmail Acocounts I have since 10+ years ago
2. Yup same problem
3. Siento que arranco la carretera, Voy enfierrado por la costera, Mi troque casi se desparpaja, Pero pura madre se me raja

What a fucking idiot.

[POf-L]

这是一个很好的收获，请分享：

1. 您使用哪种类型的电子邮件（gamil、Outlook、Temp Mail？）
2. 使用脚本和 Cursor 应用程序时，帐户的行为是否相同？例如，如果脚本在 3.7 下无法工作，Cursor App 还能在 3.7 下工作吗？

该模型的模型混淆问题并不是一个新问题。Cursor 在 2 月份使用这种策略“封禁”了异常账户，强迫异常账户使用 GPT4，并使用提示让他们认为自己是 Sonnet 3.5。我认为这可能是由于脚本的伪装策略不足以骗过 Cursor App 造成的，但还需要更多的研究。

实际上，是TLS的问题

[peachyGT]

Just noticed gpt-4o-mini or cursor's model usage on my account - which I have never used personally. I assume this is the "shadow ban" strategy kicking in - although regular cursor app functions like normal.

[JiuZ-Chn]

Share you one solution, not sure if it works in your machine:
Change the timezone to match your real timezone in this file: https://github.com/JiuZ-Chn/Cursor-To-OpenAI/blob/main/src/routes/v1.js
You need to update the line 36, line 112, line 138

[hjutahai1212]

Share you one solution, not sure if it works in your machine: Change the timezone to match your real timezone in this file: https://github.com/JiuZ-Chn/Cursor-To-OpenAI/blob/main/src/routes/v1.js You need to update the line 36, line 112, line 138

I tried, still have the same issue. Unless 'Europe/Berlin' is somehow the issue

[danyadanya99]

Share you one solution, not sure if it works in your machine: Change the timezone to match your real timezone in this file: https://github.com/JiuZ-Chn/Cursor-To-OpenAI/blob/main/src/routes/v1.js You need to update the line 36, line 112, line 138

I will cautiously say it works for me, for now at least. thank you very much!

Share you one solution, not sure if it works in your machine: Change the timezone to match your real timezone in this file: https://github.com/JiuZ-Chn/Cursor-To-OpenAI/blob/main/src/routes/v1.js You need to update the line 36, line 112, line 138

I tried, still have the same issue. Unless 'Europe/Berlin' is somehow the issue

are you using VPN? if so - timezone should match the country you are chose in VPN