Fine tuning reduces context length?

Hi, I’m using a GPT-4.1 fine-tuning output model for assigning texts to some specific categories. I tried using the regular GPT-4.1, but the results were not satisfactory. Now I’m having an issue with context length, I’m getting this error message from the API : "This model’s maximum context length is 128000 tokens. However, your messages resulted in 132119 tokens. Please reduce the length of the messages. "

I used supervised fine-tuning and from reading the documentation I thought a fine-tuned model would retain the original model’s context window (1,047,576 tokens for GPT-4.1). The only mention of a 128k context length for GPT-4.1 was for the “inference context length” on the “Fine-tuning best practices” doc, but I thought this only applied to the training data set for the fine-tuning job itself.

Is this a reading comprehension mistake on my part or did I do something wrong during the fine-tuning process ? Is there any way to get a higher context length from a fine-tuned model ?

Examples context length: “training data”, context sequence.

Inference context length: “inference”, also called sampling, generation, or calling the model.

One is how much you can train on, one is how much you have for I/O use.