Microsoft’s most succesful new Phi 4 AI mannequin rivals the efficiency of far bigger techniques

Microsoft launched a number of new “open” AI fashions on Wednesday, essentially the most able to which is aggressive with OpenAI’s o3-mini on not less than one benchmark.

All of the brand new pemissively licensed fashions — Phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 reasoning plus — are “reasoning” fashions, that means they’re in a position to spend extra time fact-checking options to complicated issues. They develop Microsoft’s Phi “small mannequin” household, which the corporate launched a 12 months in the past to supply a basis for AI builders constructing apps on the edge.

Phi 4 mini reasoning was skilled on roughly 1 million artificial math issues generated by Chinese AI startup DeepSeek’s R1 reasoning mannequin. Around 3.8 billion parameters in dimension, Phi 4 mini reasoning is designed for instructional purposes, Microsoft says, like “embedded tutoring” on light-weight units.

Parameters roughly correspond to a mannequin’s problem-solving expertise, and fashions with extra parameters typically carry out higher than these with fewer parameters.

Phi 4 reasoning, a 14-billion-parameter mannequin, was skilled utilizing “high-quality” internet knowledge in addition to “curated demonstrations” from OpenAI’s aforementioned o3-mini. It’s finest for math, science, and coding purposes, in response to Microsoft.

As for Phi 4 reasoning plus, it’s Microsoft’s previously-released Phi-4 mannequin tailored right into a reasoning mannequin to realize higher accuracy on explicit duties. Microsoft claims that Phi 4 reasoning plus approaches the efficiency ranges of R1, a mannequin with considerably extra parameters (671 billion). The firm’s inner benchmarking additionally has Phi 4 reasoning plus matching o3-mini on OmniMath, a math expertise check.

Phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 reasoning plus can be found on the AI dev platform Hugging Face accompanied by detailed technical reviews.

“Using distillation, reinforcement studying, and high-quality knowledge, these [new] fashions steadiness dimension and efficiency,” wrote Microsoft in a weblog submit. “They are sufficiently small for low-latency environments but keep sturdy reasoning capabilities that rival a lot greater fashions. This mix permits even resource-limited units to carry out complicated reasoning duties effectively.”