2026-09-06 12:07 UTC
And while the model is much larger (125b vs 27b), its limited 6b active parameters makes it faster at both prompt processing and decoding than its dense sibling.
Replies (1)
-
@prutser@mas.to 2026-09-06 12:19
Nope, the Flash-Next port is partially broken. The model probably didn't perform comprehensive end-to-end tests.