Microsoft launches Phi 4 reasoning AI models to compete with DeepSeek and OpenAI

Microsoft launches Phi 4 reasoning AI models to compete with DeepSeek and OpenAI

Microsoft has introduced a new series of AI models called the Phi-4 reasoning, designed to solve complex problems in subjects like math, science, coding, and logic. These models are part of Microsoft’s small language model (SLM) family and were built to be more efficient while still delivering high-quality results.

The most powerful model in the series is Phi-4 Reasoning, which has 14 billion parameters. It was trained using high-quality web data, including content from OpenAI’s o3-mini model. According to Microsoft, this model is their best performer when it comes to advanced tasks in math, science, and programming. It combines reinforcement learning and supervised fine-tuning, helping it reason more effectively and solve more difficult problems.

Another model, Phi-4 Mini Reasoning, is a smaller version with 3.8 billion parameters. It was trained using about one million synthetic math problems created by DeepSeek’s R1 model. Even though it is smaller in size, Phi-4 Mini is designed specifically for teaching and learning. Its lightweight design makes it easy to run on devices, making it ideal for educational tools and classroom settings.

The third model, Phi-4 Reasoning Plus, sits in between the other two in terms of size and performance. Despite being smaller than DeepSeek R1, it outperforms it on several benchmarks, especially in math. Microsoft also claims that Reasoning Plus is comparable to OpenAI’s o3-mini in terms of math problem-solving ability.

Microsoft says these models are part of a larger plan to provide smart, efficient AI that works well on everyday devices. “Phi’s evolution over the last year has continually pushed this envelope of quality vs. size,” the company said. All models in the Phi-4 reasoning family can run locally on Windows 11 devices, whether they use CPUs or GPUs, allowing users to benefit from powerful AI without needing internet access. With this launch, Microsoft continues to focus on AI that is fast, useful, and accessible, especially in the fields of education, science, and programming.

With the launch of the Phi-4 reasoning, Microsoft is taking a major step toward making advanced AI more accessible, efficient, and practical for real-world use. These models not only deliver strong performance in key areas like math, science, and coding, but also run locally on everyday devices, making them more flexible and secure. As AI continues to evolve, Microsoft’s Phi-4 models represent a promising direction toward smarter, more responsible, and user-friendly AI tools.