Phi-4-reasoning

Phi-4-reasoning is a state-of-the-art open-weight reasoning model finetuned from Phi-4 using supervised fine-tuning and reinforcement learning. Trained on a blend of synthetic and high-quality public data, it excels at math, science, and coding tasks, with a focus on advanced reasoning and alignment for safety. The model has 14B parameters and supports a 128K token context length.

Outputs include a reasoning chain-of-thought block and a summarization block. Released under the MIT license, this static model was trained on data up to March 2025. For best results, use prompts in chat format and review the license for details.