π“Ÿ host1.bitchan.one

𓀐 @?JvFzz=y

⏱ 2d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

Right now, the main problem is that any of the smaller size open weights models can't even beat GPT 6 Luna, let alone Haiku 5.5 in benches for real world performance stuff. The models that hit that threshold are all out of most people's reaches. I would like if Qwen can hit at least that but I have major doubts about this considering how the capability gap between the frontier labs and open weights has widened considerably. That is my biggest worry where Qwen 4 won't close the gap⏱ 2d↩