Phi-4 Reasoning

Small open-weight models (3.8B/14B) delivering powerful reasoning for math/science/code, rivaling larger LLMs. Available on Azure AI Foundry & HF.

Phi-4 Reasoning represents a breakthrough in small language models (SLMs), offering powerful reasoning capabilities—especially in math, science, and coding—while remaining compact and efficient. Available in three variants (Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning), these models rival larger AI systems despite their smaller size (3.8B to 14B parameters). They leverage techniques like distillation, reinforcement learning, and high-quality synthetic data to excel in tasks requiring multi-step reasoning, such as solving complex math problems or coding challenges.

Phi-4-reasoning and its enhanced version, Phi-4-reasoning-plus, outperform larger models in benchmarks, including Ph.D.-level science questions and math Olympiad qualifiers. The mini variant is optimized for low-latency environments, making it ideal for edge devices, educational tools, and mobile applications. Integrated into Windows Copilot+ PCs, these models enable fast, power-efficient AI features like real-time summarization and text intelligence.

Microsoft emphasizes responsible AI development, ensuring safety and reliability through supervised fine-tuning and human feedback. Available on Azure AI Foundry and HuggingFace, Phi-4 Reasoning models redefine efficiency in AI, balancing performance with accessibility.