A chat app resends the full conversation history on every request. By turn 60 the latency and bill have both roughly tripled, though each new user message is short. What's happening?

Answer: B — The API is stateless — your app resends all prior turns, so input tokens grow with every turn

Statelessness is the root cause of cost creep in long chats. The model holds nothing between calls; your application resends history, so turn 60 pays for 60 turns of input. The fix is compaction or pruning, not a different model.