Another deepseek moment? it seems they have fully caught with fable tier of models, and this was a lot sooner than was expected.
Another deepseek moment? it seems they have fully caught with fable tier of models, and this was a lot sooner than was expected.
Yeah, I would have expected Zhipu to ship a Fable-adjacent model by the end of the year, but the jump from Kimi 2.7 (which I think is just barely at the level where it is genuinely helpful for coding) to this is absolutely bonkers. And this is clearly not just benchmaxing; this thing actually works.
If you told me I could only use this and never use Fable or Sol again, I'd shrug and not feel like I'd lost much.
Now it seems the best way to tell if a frontier model is benchmaxxed is to check if it can autonomously solve a major open mathematical problem.
> Yeah, I would have expected Zhipu to ship a Fable-adjacent model by the end of the year
There were talks of a GLM 5.3 in August, so maybe not that far away...