sanitation@lemmy.today to Technology@lemmy.worldEnglish · 6 days agoOpenAI made $13 billion in 2025 and lost $21 billion doing itwww.techspot.comexternal-linkmessage-square95fedilinkarrow-up1429arrow-down14
arrow-up1425arrow-down1external-linkOpenAI made $13 billion in 2025 and lost $21 billion doing itwww.techspot.comsanitation@lemmy.today to Technology@lemmy.worldEnglish · 6 days agomessage-square95fedilink
minus-squaremasterspace@lemmy.calinkfedilinkEnglisharrow-up2arrow-down1·edit-25 days agoNo they’re not. The best Chinese models are good, but they’re resoundingly outclassed by the likes of Claude. And that’s not to mention the Chinese models are trained off the American models, i.e. they can’t exist without the American ones being developed first.
minus-squaresanitation@lemmy.todayOPlinkfedilinkEnglisharrow-up1arrow-down1·edit-24 days agoThey don’t train off American models, that’s just their PR narrative to explain losing the race to a country without latest hardware
minus-squaremasterspace@lemmy.calinkfedilinkEnglisharrow-up1·3 days agoThey’ve literally been caught using distillation training off the American models. They undoubtedly do. Even the more efficient American models do.
minus-squaresanitation@lemmy.todayOPlinkfedilinkEnglisharrow-up1arrow-down1·4 days agohttps://m.youtube.com/watch?v=-DVxMLFhi3o Glm 5.2 is almost the same as fable, difference is immaterial. I use deepseek v4 pro daily and it’s fine for most tasks and so cheap that I stick those deepseek api keys into most of my applications.
No they’re not.
The best Chinese models are good, but they’re resoundingly outclassed by the likes of Claude.
And that’s not to mention the Chinese models are trained off the American models, i.e. they can’t exist without the American ones being developed first.
They don’t train off American models, that’s just their PR narrative to explain losing the race to a country without latest hardware
They’ve literally been caught using distillation training off the American models. They undoubtedly do. Even the more efficient American models do.
https://m.youtube.com/watch?v=-DVxMLFhi3o Glm 5.2 is almost the same as fable, difference is immaterial.
I use deepseek v4 pro daily and it’s fine for most tasks and so cheap that I stick those deepseek api keys into most of my applications.