Hidden Gem
#1
by singulariti - opened
This model is incredibly good for the size, yet the thinking JSON structure is slightly broken; the formatting causes most CLIs to not understand when the model is thinking compared to generating its response or tool calling. Yet I can't believe the near Sonnet 4.5 performance!
This comment has been hidden (marked as Off-Topic)
This model is incredibly good for the size, yet the thinking JSON structure is slightly broken; the formatting causes most CLIs to not understand when the model is thinking compared to generating its response or tool calling. Yet I can't believe the near Sonnet 4.5 performance!
I’m using it too, and so far I haven’t run into any obvious bugs. Do you have a specific case where the thinking JSON/output formatting breaks parsing or causes problems in the CLI?