凳都未坐暖，Alibaba又出左個號稱超越Deepseek V3嘅Qwen2.5-Max

Qwen2.5-Max (MoE)
https://x.com/alibaba_qwen/status/1884263157574820053?s=46

暫時未係open weight

俾少少 background：
Qwen2.5/Qwen2.5 Coder一直喺Deepseek V3出之前，open weight LLM裏面最高分嘅model。Chinese LLM冒起其實冇好多人眼中咁突如其來，起碼都發展左一年有多，huggingface 上面一路都有得 download來用。有跟開就知過去半年open weight model已經仰中國佬嘅世界

美國佬除左meta之外都係 closed models，一就係Gemma/Phi哩類虧好多嘅細model，歐洲佬個mistral半收皮，印度佬日本仔一條毛都冇
我都唔鐘意共産黨，不過諷刺地，過去半年open weight model全靠中國佬 model頂住

死鬼佬班digital nomad點同我地996 it9鬥

係時候整條鯰魚嚟炸炸班死鬼佬啦

Open source 萬歲，closed source 慢慢自high啦

冇放weight，咁不如繼續用chatgpt

似乎Qwen嘅max series一向唔係open weight，唔知Deepseek嘅出現會唔會改變佢地嘅諗法？

因為qwen用落好廢咪無人care
但DS用落係勁