2026-10-18 · America/Los_Angeles · 社区动态 · #13

Turning off "thinking" on open models over API: what each one accepts as of October 2026

reasoning models are awesome until all you want back is a Json objet but instead get 3k tokens of 'wait, hold on, let me reconsider first'. thinking bill as output, the eat into max_tokens and on a lot of routine agent steps they add no thing. so turning off thinking for boring calls is a easy win win Altho except every model family handles it different now and a couple of the newest ones dropped the off switch completely, a comparison illustration given below as per the official docs- Model Can you turn thinking off? How Effort levels (default) Qwen 3.6 / 3.8 Yes chat_template_kwargs:…

热度 45.3 / 100；排名与评分保留该期记录。