Phi-4 Reasoning
Phi-4 Reasoning is the reasoning-focused 14B descendant of Phi-4. Its 32K context and response format make it a separate integration from the original text checkpoint.
What distinguishes this release
The card describes training on reasoning traces covering mathematics, science and coding. It documents a reasoning portion followed by an answer portion in generated responses.Phi-4 Reasoning official model card
Build the task around its input and output
Practical guidance. Use the final answer as the application result and retain only the explanation needed by the user. A reasoning trace is not independent evidence. For a code or calculation task, rely on observable output and explicit checks of the result.
Review the result in its intended setting
Practical guidance. Keep the input focused on one problem and define what counts as completion. Compare it with the Plus release using the full response duration and verified answer quality. The shared 14B size alone does not determine their operating cost.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | microsoft/Phi-4-reasoning |
| Context window | 32,768 tokens Published checkpoint context; serving engines may impose lower limits Phi-4 Reasoning official model card |
| Model size | 14B densePhi-4 Reasoning official model card |
| Input and output | Text input and text outputPhi-4 Reasoning official model card |
| Weight license | MITPhi-4 Reasoning official model card |