GLM 5.2 increased its CritPt score over GLM 5.1 by 4.5 times, mirroring Opus 4.5 => Opus 4.8 evolution (except it took 2 months, not 7). DeepSeek V4.1 merely doubling over V4 would be solidly GPT 5.5-high territory. I think they can do it.
We're entering a strange territory.
This thing is an absolute monster. Everyone was waiting for the next DeepSeek event, but it has arrived under a different name.
https://t.co/k9jOQSQLr3
GLM 5.2 is now on DeepSWE as the top open-source model on our leaderboard.
With a pass@1 score of 44% at max effort, GLM 5.2 is indisputable #1 open-source model besting Kimi K2.7 Code by 17%.
@NasheedGroyper it seems not bad (thought it was like madoka magca type.. uggh)
Fr, I thought apple eating an apple.. huh any regret ?
Nah it's not, which is disappointing
Dario's answer to Dwarkesh on why he won't buy $300B-1T of compute sure was incomplete. But I guess it would have been unwise to spell out "our competitors will willingly surrender their clusters because *they* have overbuilt and can't generate revenue with their shit models"