Microsoft (Phi)
Small Language Models proving that data quality outperforms raw parameter scale.
About
About Microsoft (Phi)
Microsoft Research's Phi model series demonstrates that carefully curated training data produces models that punch far above their parameter weight. Phi-4 (14B parameters) outperforms many 70B models on reasoning benchmarks, while Phi-3.5 Mini (3.8B) runs efficiently on-device. The Phi series is especially popular for edge deployments, mobile AI, and research into the scaling laws of high-quality synthetic data.
Key highlights
Phi-4 outperforms models 5× its size on GPQA Diamond and MATH benchmarks
MIT licence — free for commercial use, modification, and redistribution
Phi-3.5 Mini targets on-device inference for Windows Copilot+ and mobile
Wide availability on Azure AI Studio, Ollama, llama.cpp, and Hugging Face
Research-grade focus on synthetic data quality rather than raw scale
Models
3 models tracked
Context windows, pricing, capabilities, and API identifiers — everything you need to choose the right model for your workload.
Phi-4
2024
Microsoft's latest flagship small model. At 14B parameters, Phi-4 outperforms many much larger models on GPQA and MATH benchmarks thanks to high-quality synthetic training data and careful curriculum.
- Outperforms 70B models on reasoning
- MIT licence
- 14B parameters
- High-quality synthetic data
Phi-4Phi-3.5 Mini
2024
A 3.8B model with a 128K context window that runs efficiently on CPU and NPU hardware. Designed for device-side AI in Windows Copilot+ PCs and mobile applications.
- On-device inference
- 3.8B parameters
- 128K context
- CPU & NPU optimised
Phi-3.5-mini-instructPhi-3.5 MoE
2024
A mixture-of-experts variant of Phi-3.5 that adds vision capability. Uses 16 experts with 2 active per token for efficient inference at competitive quality.
- Vision capable
- MoE efficiency
- 128K context
- Open weights
Phi-3.5-MoE-instructUse Cases
What Microsoft (Phi) is best for
Each provider has specialisms shaped by their architecture decisions, training data, and API design. Here is where Microsoft (Phi) performs best.
Get started
Start building with Microsoft (Phi)
Visit the official Microsoft (Phi) platform to access API keys, documentation, and pricing details.