
Qwen3.8-27B: how much reasoning effort should we pay for?
The new Qwen3.8 lets us tell it how much thinking effort to use before it answers. More thinking should buy better answers, and it certainly buys longer waiting time, so the question is how much is worth paying for. We put Qwen3.8-27B through three effort settings against two other models we regularly use. The metric we use is Substance, which is the mean of accuracy and completeness.

























