What's this about?
Mollick says K3 and GLM-5.3 are good open-weight releases but feel substantially behind Mythos and Astra, making the current gap larger than it has been for some time.
What exactly are “open weights”?
They are downloadable model parameters that organizations can run and often adapt on their own infrastructure. This grants control and privacy, but does not necessarily reveal training data or make the whole system open source.
Why is this worth noting?
The post drew about 367 likes and 35 replies. It is useful practitioner signal for deployment choices, but remains one user's judgment rather than a controlled evaluation; Mollick expects the gap to change again.