Over the past few days, I’ve noticed that the default “Thinking” model selected in Cursor’s Auto mode seems to perform with significantly lower accuracy than before. The responses often begin with phrases like “Hello! As an AI pair programmer, I’ve analyzed the issue you’re facing,” which feels reminiscent of older LLM styles.
Because of this, I’ve started manually switching out of Auto mode depending on the task. For example, I use ChatGPT 4o when I want to code quickly, and switch to models like “o4 mini” or “gemini-2.5-pro-preview-05-06” when I anticipate a more complex problem.
I’m still evaluating which model provides the most reliable results overall. Do you have any recommendations or preferred models you find particularly effective?
Thanks in advance for any insights!