Claude 3.7-thinking worse than claude 3.7?

The claude 3.7 model performs tasks much more accurately and faster than the 3.7-thinking, or is it just me?

It depends a lot on the prompt, project, rules, requirements etc.
With unstructured task and without rules its easier to go wrong as it then starts making assumptions. So if you are not well familiar with how to prompt thinking models differently from regular it might be easier to use 3.7 or even 3.5.

Can you share what are you trying to do?