They already release one before but I was not super impressed tbh.
However, if they take this top model and make it with reasoning, then I would expect a huge leap. Open Source community is definitely eating good this and next year for sure
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
However, if they take this top model and make it with reasoning, then I would expect a huge leap. Open Source community is definitely eating good this and next year for sure
Nope they never released the weights, only qwen did with qwq, and the r1 lite preview was 16b to 32b.... a 685b would be way better
so far they released the base and instruct tuned of 685b v3 but no model cards yet and no official announcememt. i hope they release a reasoning version today that surpasses o1
a 685b reasoning model would actually put the data center on fire lol
But yeah, definitely smart that they made it MoE, looks like a nice architecture for reasoning given the huge token output need
btw, I would find that claim to be unlikely.
I'm sure they will get there. After all, these Chinese shops still just have less resources.
5 million was the cgi budget of Terminator 2.... so for the price of fake graphics in 1994, 3 decades later China can get quasi AGI
Strong results from Chinese companies continues without pause. I think 2025 is going to be pretty crazy in terms of AI progress across the board.
