2 ms·

You can always implement summarization, langchain for example support this. But that's not changing how many tokens the model itself is consuming