Hey, thanks for the report and the Request IDs, I can see the screenshot with garbled text.
This is a known issue with the model behind Auto mode. It sometimes returns broken Unicode output, especially for non-English languages. The team is aware, but we don’t have an ETA for a fix yet.
As a workaround, try switching from Auto to a specific model, like Claude Sonnet or GPT. This should avoid the issue since it’s tied to Auto routing to a particular model.
Hi, thanks for your response. I did switch to the GPT-5.4 model and it worked well, but choosing a specific model would result in heavy API token consuming while the Auto capacity is left unused.
It used to work very well until last week, which led me to upgrading. Please consider adding this issue to your fix plan soon.
Hey, good news: this specific issue where Auto mode produced garbled Unicode has been fixed since your original report. The fix rolled out a few days ago.
Can you try switching back to Auto mode and see if Vietnamese now renders correctly? If it’s still broken, send me a new Request ID and your current Cursor version so I can dig in.
About token usage: when you use Auto, requests come from your included Pro quota, so you shouldn’t worry about extra API costs. Specific models like GPT-5.4 also use your included requests unless you’re on Max mode or using your own API key.
