The number of tokens send over to the server is too small. Currently, the number of characters to be generated is subtracted from the maximum allowed (2028). For example, with 500 characters, only a total of 1528 tokens are transmitted.

But with the general simplification of 1 token = 4 characters, it should be 2028 - 500/4 = 1903 tokens instead. That is a noticeable difference (in this instance, 375 lost tokens)