alt Hacker NewsWhy can we say fable 5.2 is better than 5.1 if it needs double the tokens to complete the task? Somehow shouldn't it be compared at same token usage level to see which one is smarter? Why is not the case?
Why can we say fable 5.2 is better than 5.1 if it needs double the tokens to complete the task? Somehow shouldn't it be compared at same token usage level to see which one is smarter? Why is not the case?
Reasoning tokens are a path. If your world is small, a small number of steps may suffice.