Microsoft’s most capable new Phi 4 AI model rivals the performance of far larger systems

Microsoft various new “open” AI models launched On Wednesday, the most capable of these is competitive with OpenAi’s O3-Mini on at least one benchmark.

All new pemissively licensed models-Phi 4 mini-reasoning, Phi 4 reasoning and Phi 4 Reasing plus-being “reasoning” models, which means that they can spend more time checking facts control solutions for complex problems. They are expanding the Phi “Small Model” family from Microsoft, which launched the company a year ago to provide a basis for AI developers who build apps on the edge.

Phi 4 Mini reasoning was trained at around 1 million synthetic math problems generated by the Chinese AI startup Deepseek’s R1 Redeneer model. About 3.8 billion parameters in size, Phi 4 Mini reasoning is designed for educational applications, says Microsoft, such as “embedded tutoring” on lightweight devices.

Parameters are roughly in line with the problem -solving skills of a model and models with more parameters generally perform better than those with fewer parameters.

Phi 4 reasoning, a model of 14 billion parameter, was trained with the help of “high-quality” web data and “composite demonstrations” of OpenAi’s previously mentioned O3-Mini. According to Microsoft, it is best for mathematics, science and coding applications.

Regarding Phi 4 Reasing Plus, it has been adapted to the previously released Phi-4 model of Microsoft in a reasoning model to achieve better accuracy at certain tasks. Microsoft claims that Phi 4 reasoning plus the performance levels of R1 approaches, a model with considerably more parameters (671 billion). The internal benchmarking of the company also has Phi 4 reasoning plus the matching of O3-Mini on Omnimath, a mathematical skills test.

Phi 4 MINI REEDING, PHI 4 REEDING AND PHI 4 REASURE PLUS are available on the AI DEV -Platform Cuddling Face accompanied by detailed technical reports.

“With the help of distillation, reinforcement education and high -quality data, this, this one [new] Models Balance size and performance, ”wrote Microsoft in a Blog post. “They are small enough for environments with low latency, but nevertheless retain strong reasoning options that compete much larger models. With this blend, even devices with a resource -limited devices can efficiently perform complex reasoning tasks.”