Microsoft's Phi models, such as Phi-4, are small language models that excel at complex reasoning in areas like math, science, and coding, offering high performance in a compact size.
A 530B-parameter Transformer developed by NVIDIA and Microsoft, one of the earliest models to cross 500B parameters:contentReference[oaicite:33]{index=33}.
View details →A 1.3B-parameter model trained on curated high-quality data, demonstrating strong performance despite its small size:contentReference[oaicite:34]{index=34}.
View details →A 2.7B parameter follow-up model focusing on textbook-quality data, used to explore data efficiency:contentReference[oaicite:35]{index=35}.
View details →A 14B-parameter model (2024) marketed as a "small language model" by Microsoft, emphasizing efficiency.
View details →