Phi-3 Medium 128K

Phi-3 Medium 128K is Microsoft's 14-billion-parameter mid-tier model in the Phi-3 family, released May 2024 with a 131K-token context window and an MIT license. Microsoft trained it on heavily filtered, textbook-quality synthetic data — the bet being that data quality at small scale beats raw parameter count. It delivers MMLU scores and GSM8K accuracy that exceed most 14B peers, closing some of the gap to 70B-class models on reasoning-heavy tasks. The 131K context makes it viable for long-document summarization and extended code reviews without chunking. Hosted coverage skews toward Azure AI and a handful of open providers; if you need the breadth of Llama or Qwen's provider ecosystem, that is currently a real constraint.