Microsoft has released three new AI reasoning models Phi 4 that allow you to perform complex tasks even on weak devices
Microsoft has introduced several new open AI reasoning models Phi 4 that can compete with OpenAI o3-mini.
Microsoft has introduced several new open AI reasoning models Phi 4 that can compete with OpenAI o3-mini.
All three models — Phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 reasoning plus — are “reasoning” models, meaning they spend more time checking facts when solving complex problems. They expand Microsoft’s Phi “small model” family, which the company launched a year ago, TechCrunch reports .
Phi 4 mini reasoning
This model was trained on about 1 million synthetic math problems generated by the R1 reasoning model of Chinese startup DeepSeek. The Phi 4 mini has about 3.8 billion parameters and is designed for educational applications such as “embedded learning” on lightweight devices.
Phi 4 reasoning
Phi 4 reasoning has 14 billion parameters. It was trained on high-quality web data and carefully selected examples from OpenAI's o3-mini. According to Microsoft. It is best suited for tasks related to math, science, and programming.
Phi 4 reasoning plus
This is the previously released Phi-4 model, adapted into a reasoning model to achieve better accuracy. Microsoft claims that Phi 4 reasoning plus approaches the performance level of R1, a model with a significantly larger number of parameters (671 billion). In the company's internal benchmarking, Phi 4 reasoning plus also matches the o3-mini on OmniMath, a test of mathematical skills.
Phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 reasoning plus are available on the Hugging Face AI development platform along with detailed technical reports.
“Using distillation, reinforcement learning, and high-quality data, these models balance size and performance,” Microsoft wrote in a blog post. “They are small enough for low-latency environments, yet retain powerful reasoning capabilities that rival much larger models. This combination allows even resource-constrained devices to efficiently perform complex computational tasks.”
Recall that Xiaomi recently launched its first AI model, MiMo . This is an open-source reasoning model that is fully trained by the company itself.
Microsoft has created a small AI model, Phi-4, that solves mathematical problems better than much larger models from Google
Microsoft has created a small AI model, Phi-4, that solves mathematical problems better than much larger models from Google
"As powerful as GPT-3.5, only in a smaller form factor." Microsoft introduced a lightweight AI model for smartphones — Phi-3 Mini
"As powerful as GPT-3.5, only in a smaller form factor." Microsoft introduced a lightweight AI model for smartphones — Phi-3 Mini
Read the country's main IT news in our Telegram
Read the country's main IT news in our Telegram
