Microsoft has unveiled a new suite of open-source AI models under its Phi-4 series, aiming to deliver powerful reasoning capabilities in compact sizes. The lineup includes Microsoft’s Phi-4 Mini Reasoning, Phi-4 Reasoning, and Phi-4 Reasoning Plus, each designed to handle complex tasks like math, coding, and scientific problem-solving, even on devices with limited resources.
Phi-4 Mini Reasoning, with 3.8 billion parameters, was trained on approximately 1 million synthetic math problems generated by DeepSeek’s R1 reasoning model. It’s tailored for educational applications, such as embedded tutoring on lightweight devices.
Phi-4 Reasoning, a 14-billion-parameter model, utilizes high-quality web data and curated demonstrations from OpenAI’s o3-mini. It’s optimized for math, science, and coding applications.
Microsoft’s Phi-4 Reasoning Plus is an adaptation of the previously released Phi-4 model, enhanced through reinforcement learning to achieve better accuracy on specific tasks. Microsoft claims that Phi-4 Reasoning Plus approaches the performance levels of DeepSeek’s R1 model, which has significantly more parameters (671 billion). Internal benchmarks also show it matching o3-mini on the OmniMath test.
All three models are available on platforms like Hugging Face, accompanied by detailed technical reports. Microsoft emphasizes that these models balance size and performance, making them suitable for low-latency environments while maintaining strong reasoning capabilities.
Source: (Techcrunch)
