I am doing simple NLP, parsing 20 or so fields from user text into a JSON object. I am using the create response API call. This works fine with both gpt4 and gpt5. However, I notice that the reasoning level (for gpt5) does not make much difference. For example, effort = low does as well as effort = medium. Medium just takes three times longer. So here is my question: For NLP, does reasoning in general get you anything? Or is it not needed (much). Thanks in advance for your help.
Does your task take advanced knowledge, problem-solving or decision-making. Doesn’t really sound like it.
Have a click here: AI Model & API Providers Analysis | Artificial Analysis
However, benchmarks can’t give you results for your particular application. For example, a “simple” code refactor task earlier today, just over and over, restructuring the input knowledge and developer message, gpt-5 was just terrible. Worse is: you try another typical “chat” turn with corrections and the code is filled with dumb comments that don’t believe facts or your input, are chatting about the specific task, and don’t take ownership for the prior garbage assistant output.
Had o3 do it.
Thanks for the response! I appreciate the help.
yeah master octopus is definitely the Ai tamer