Token Budget / One Request, GPT-5.4 Nano
1,500 tokens total  /  1,000 input + 500 output
SYSTEM
HISTORY
RETRIEVED
OUTPUT
System prompt 300 tk
History 400 tk
Retrieved 300 tk
Output 500 tk
Cost per request
input @ $0.20/1M  /  output @ $1.25/1M
$0.000000
Output tokens cost 6.25x more than input. Cap max_tokens and the cheap segment stays cheap.