Phi-4 Reasoning Plus
Phi-4 Reasoning Plus adds reinforcement learning to the Phi-4 reasoning line. Microsoft reports longer responses as part of the trade-off, so it should be assessed on complete task cost and latency.
What distinguishes this release
The model card retains a 32K context and the 14B dense architecture. It reports roughly 50% more generated tokens on average than the preceding reasoning model; this is the provider's observation, not an independent measurement.Phi-4 Reasoning Plus official model card
Build the task around its input and output
Practical guidance. Use the Plus release when extra reasoning changes the correctness of difficult cases in your own workload. Compare it on the same questions, stopping rules and result criteria as Phi-4 Reasoning. Record how often the extra output leads to a useful correction.
Review the result in its intended setting
Practical guidance. For interactive work, budget for the whole answer rather than the first visible token. If the user only needs a short decision, request the result and the minimum supporting explanation. A larger response does not by itself provide a better decision.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | microsoft/Phi-4-reasoning-plus |
| Context window | 32,768 tokens Published checkpoint context; serving engines may impose lower limits Phi-4 Reasoning Plus official model card |
| Model size | 14B densePhi-4 Reasoning Plus official model card |
| Input and output | Text input and text outputPhi-4 Reasoning Plus official model card |
| Weight license | MITPhi-4 Reasoning Plus official model card |