Post #2200262
2026-05-06 20:18 UTC
If I want something done well and as fast as possible, I shouldn't be offering more money the more time this thing takes. We can assume that hourly pricing isn't literally about the time, hours are just a unit for negotiation.
Token pricing for LLM coding is similar to this. Tokens are there for the benefit of the agent, their tool calls, chain-of-thought and overall context.
The latest 5.5 #OpenAI models question the hourly pricing approach by trying to do the thing right with less tokens.
Replies (1)
-
@t_var_s@phpc.social 2026-05-07 09:18
It's a nice goal to have, but without better discernment, the models just try to reduce scope as much as possible to deliver the bare minimum of what was prompted. Put the price difference over that and 5.4 may still offer better usage, even when it's wordy for its own sake.