True. Also, the APIs don't care too much about restricting output length, they might actually be more verbose to charge more

It's interesting how the same model being served through different interfaces (chat vs api), can behave differently based on the economic incentives of the providers