pull down to refresh
I think they'll publish because otherwise they need to think up a pretty good reason to safe face.
Official reason "Dario is right, issa verrry dangerous uwu"
Actual reason: "There is so much money in closed weights and undercutting the competition"
reply
oh... that's what you can tell to Americans whom would accept it and ask where to insert credit card - brrrrrrr.
I have difficulty imagining this being received well internally though, unless the narrative would be: "use GLM, it is better and free. We are now targeting foreigners to milk some moneys from their printer"
reply
I think they'll publish because otherwise they need to think up a pretty good reason to safe face. Honor culture works in the favor of the open weights consumer right now.
Secondly, although their own benchmarks put it at the top, it scores lower on other's benchmarks[1][2] vs both Fable and Sol and it's hard to judge usablity - they admitted as much as inferiority themselves in their publication. If they close up now, they lose the next iteration also - like Meta. So it'd be a huge risk.
I only had some low-end tasks for it thus far so I cannot comment on it yet, but I'll mix it into my upcoming real-world comparison next to GLM-5.2, Deepseek-4-Pro and Qwen-3.6-Max, where I baseline against Opus 4.8 performance.
https://arena.ai/leaderboard/text ↩
https://artificialanalysis.ai/leaderboards/models ↩