Microsoft Open Sources Phi-4 Reasoning Models

Microsoft has open-sourced its new Phi-4 family of reasoning models, designed to compete with leading AI models while remaining lightweight enough to run on consumer devices like laptops and smartphones. The release includes three variants—Phi-4 Reasoning, Phi-4 Reasoning+, and Phi-4 Mini-Reasoning—each optimized for different performance and hardware needs.

Phi-4-Reasoning is a 14B-parameter model designed for complex reasoning tasks. Trained through supervised fine-tuning on carefully curated “teachable” prompts and o3-mini-generated reasoning demos, it produces detailed reasoning chains that efficiently use inference-time compute.

Whereas Phi-4-Reasoning+ is an enhanced variant using outcome-based reinforcement learning to generate longer, more accurate reasoning traces. Both models outperform much larger open-weight models like DeepSeek-R1-Distill-Llama-70B and approach the full DeepSeek-R1's performance across math, science, coding, and planning tasks.

Similarly, "Phi-4-mini-reasoning is a lightweight open model built upon synthetic data with a focus on high-quality, reasoning dense data further finetuned for more advanced math reasoning capabilities. The model belongs to the Phi-4 model family and supports 128K token context length," Microsoft said.

All the models are now available on Hugging Face.