Phi 3 medium 128k instruct

Phi 3 Medium punches above its weight class for a 14B model, reflecting Microsoft's research focus on training efficiency over raw scale. It handles long-context tasks with a 128k token window, making it comfortable with lengthy documents or extended conversations. The trade-off is that it can occasionally struggle with complex multi-step reasoning where larger models have a clear edge.