Microsoft Unveils New Phi 4 AI Models Targeting Enhanced Reasoning Capabilities

Microsoft unwrapped three new AI models—Phi 4 mini reasoning, Phi 4 reasoning and Phi 4 reasoning plus. developer platform Hugging Face. These new models, specifically tuned to exhibit more advanced reasoning capabilities, are intended to tackle multifaceted problem-solving endeavors with greater precision and factual verification.

The Phi 4 reasoning model Wow’s with 14 billion parameters. It was trained on high-quality internet data and hand-picked examples from OpenAI’s o3-mini. This deep training further improves its capacity to answer intricate questions. Consequently, it is an indispensable tool equally powerful for developers and researchers. Phi 4 mini reasoning has a whopping 3.8 billion parameters. It’s purpose-built for educational use cases, providing embedded tutoring on devices that are as portable as they are powerful.

PHI 4 mini reasoning was trained on about 1 million synthetic math problems produced by DeepSeek’s R1 reasoning model. Such an emphasis on use in education further highlights Microsoft’s efforts to promote learning through technology. The model intentionally aims to spend more time verifying proposed solutions. This greater emphasis on precision is incredibly important in educational environments.

Phi 4 reasoning plus Builds on Microsoft’s previous Phi-4 model. Since then it has been adapted into a much more specialized model for this specific style of reasoning. This particular adaptation markedly improves performance on task-oriented dialogue data. It gives o3-mini a run for its money in standardised test environments such as OmniMath, a comprehensive math proficiency test.

Along with these releases, Microsoft has released very detailed technical reports, explicating the architectures and training methodologies for each of the models in-depth. The e-company underlined how important these advances are, calling

“Using distillation, reinforcement learning, and high-quality data, these [new] models balance size and performance.” – Microsoft

The launch of these models marks the biggest advancement in AI technology since the start of this craze. They are more powerful than ever at reasoning and problem-solving. And through a combination of the best, most diverse datasets, with advanced and cutting-edge training technologies, Microsoft is going for the AI performance gold.