70B评分吊打各家商用大模型 
主要是将类似COT的思路放在了模型输出的自定义的标签里 
佬儿们怎么看？ 
 [image] 
链接: 
https://x.com/mattshumer_/status/1831767014341538166 
https://www.reddit.com/r/LocalLLaMA/comments/1f9um6s/excited_to_announce_reflection_70b_the_worlds_top/