Microsoft launched some new models of he “open” on Wednesday, the most capable of which is competitive with Openi’s O3-Minini in at least one landmark.
All new land-licensed models-phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 Reasoning Plus-are “reasoning” models, means that they are able to spend more time to control the facts on complex problems. They expand Microsoft’s PHI “Small Model” family, which the company launched a year ago to provide a foundation for developers of the one building in Edge.
Phi 4 mini reasoning was trained for about 1 million synthetic mathematical problems created by the Chinese beginning of that model of the Deepseek reasoning. About 3.8 billion parameters in size, Phi 4 mini reasoning is created for educational applications, says Microsoft, as “embedded tutoring” on light devices.
The parameters correspond to approximately the ability to solve the problem of a model, and models with more parameters generally perform better than those with fewer parameters.
Phi 4 reasoning, a model with 14 billion parameters, was trained using online data “high quality”, as well as “curated demonstrations” from Openai’s aforementioned O3-MINI. Better is best for applications for mathematics, science and coding, according to Microsoft.
As for the Phi 4 reasoning Plus, is the PHI-4 model released by Microsoft adapted to a reasoning model to achieve better accuracy in special tasks. Microsoft claims that PHI 4 Plus reasoning approaches R1 performance levels, a model of parameters significantly more (671 billion). The company’s internal benchmarking also has a Phi 4 Plus reasoning that matches O3-Mini in Omnimath, a mathematical skill test.
Phi 4 Mini Reasoning, Phi 4 Reasoning and Phi 4 Plus Reasoning are available on the AI DeV platform embracing the face accompanied by detailed technical reports.
Techcrunch event
Berkeley, ca
|
June 5
Reserve now
“Using distillation, learning of reinforcement and high quality data, these (new) models of balance and performance,” Microsoft wrote in a blog post. “They are small enough for low latency environments, but keep strong reasoning skills that rival much larger models. This mix allows even limited resource equipment to perform complex tasks efficiently.”
