Microsoft Announced New Phi-4 AI Model

On Wednesday, April 30th, Microsoft launched new Phi-4 AI Model.

One year ago, Microsoft introduced small language models (SLMs) to customers with the release of Phi-3 on Azure AI Foundry.

Phi-3 refers to a series of small language models (SLMs), as part of their Phi model family. These models are designed to offer high reasoning and language understanding capabilities while being much smaller and more efficient than large-scale models.

Microsoft’s release of Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning signals a new chapter for small language models (SLMs) by setting a new standard for what can be achieved with small and efficient artificial intelligence systems.

What is “Reasoning model”?

Reasoning models are a type of artificial intelligence (AI) designed and trained to perform complex tasks by breaking them down into smaller steps and understanding relationships between these steps. Reasoning models require logical thinking and decision-making ability in order to generate solutions that involve multiple layers of thought or steps.

Key Features of Reasoning Models:

• • • • Logical Processing: They can apply logical rules to draw conclusions from given information. This includes tasks like breaking down a complex task into smaller and specific tasks (deductive reasoning), detecting patterns in a set of input data such as images , text ,sounds, etc.. (pattern recognition), and identifying cause-and-effect relationships (causal reasoning).
      • Multi-step Problem Solving: capability to break down a complex problem into smaller steps, processing each step, and then combining them to reach a solution.
      • Knowledge Representation: AI Reasoning Models use structured formats to store and organize information.
      • Inference Mechanisms: These are the algorithms and techniques that AI models use to get new knowledge or conclusions from the represented knowledge.
      • Contextual Understanding: They can maintain and use context over time, which allows them to make more accurate decisions based on a series of events or steps, rather than isolated data points.
      • Reflection and Adjustment: Some reasoning models can reflect on their own processes (called internal reflection) and adjust their approach to refine their decision-making, especially if they encounter uncertainty or new information.

Reasoning Models excel in mathematical reasoning and are becoming fundamental to agent-based applications that involve complex, multi-dimensional tasks.

Now going back to Phi-reasoning models, on a Microsoft blog is written that: ” Phi is a new version of SLMs (small language model) which balance the size and performance by using distillation, reinforcement learning, and high-quality data. Phi models have become an integral part of Copilot+ PCs with the NPU-optimized Phi Silica variant. This highly efficient and OS-managed version of Phi is designed to be preloaded in memory, and available with blazing fast time to first token responses, and power efficient token throughput so it can be concurrently invoked with other applications running on your PC. Despite their significantly smaller size, both models achieve better performance than OpenAI o1-mini and DeepSeek-R1-Distill-Llama-70B at most benchmarks, including mathematical reasoning and Ph.D. level science questions. They achieve performance better than the full DeepSeek-R1 model (with 671-billion parameters) on the AIME 2025 test, the 2025 qualifier for the USA Math Olympiad. Both models are available on Azure AI Foundry “