Microsoft launches Phi-4-Reasoning-Plus, a small, powerful, open weights reasoning model!
1 min read
Summary
Microsoft Research has developed a new AI model, Phi-4-Reasoning-Plus, optimized for use in applications that require deep, structured reasoning, such as mathematics, science and coding.
It is a 14-billion parameter model based on the transformer architecture and uses 16 billion tokens pulled from both synthetic and web-based datasets for training, reinforced by a further 6,400 math-related problems.
The model has been released under a permissive licence and offers flexible deployment across a range of widely-used inference frameworks and provides detailed system prompt recommendations for developers.
Crucially, Phi-4-Reasoning-Plus outperforms larger, open-weight models and is designed to deliver high-quality reasoning under memory or latency constraints, such as in a chat or embedded device, while also offering extensive safety testing and use guidelines.
Its development showcases Microsoft’s growing focus on smaller models that can rival the performance of much larger systems, and it is positioned as a research tool and component for generative AI systems rather than a turnkey solution.