Microsoft
“Small” models trained on carefully curated data: a lot of reasoning for their size.
Phi-4 reasoning
April 2025
Phi-4
December 2024
Phi-3.5
August 2024