Phi
by Microsoft · azure.microsoft.com ↗
Family of small open-weight language models from Microsoft, designed for strong reasoning at compact sizes.
open-weights small-language-model llm
- Category
- Open Models
- Business model
- Open (open weights/source)
- Availability
- Global
- Launched
- 2023
- Record updated
- 2026-07-30
- Model family
- Phi
- License
- MIT
- Founded
- 1975
- Headquarters
- US
- Canonical URL
- https://globalaiproductindex.com/products/phi/
Overview
Phi is Microsoft's family of small open-weight language models designed to deliver strong reasoning at compact sizes, from a few billion parameters up to Phi-4's 14B. Trained heavily on curated synthetic data, Phi models suit on-device and edge deployment. They are released under the MIT license and are free to use, including commercially.
Key features
- Small models with outsized reasoning performance
- Sizes suited to on-device and edge inference
- Multimodal and mini variants
- MIT-licensed open weights
Use cases
- Running capable models on laptops and phones
- Cost-efficient inference for simple tasks
- Research on synthetic-data training
Pricing
Phi models are free under the MIT license; hosted inference through Azure AI Foundry is billed by usage.
Frequently asked questions
What is Phi?
Phi is Microsoft's line of small language models that punch above their size in reasoning, math, and coding.
Are Phi models open?
Yes, Phi models are released with open weights under the permissive MIT license.
Where can Phi models run?
Locally on consumer hardware, on edge devices, or as paid hosted endpoints on Azure.
Similar products
- Baichuan — Open-weight bilingual LLM family from Beijing-based Baichuan Intelligence, now focused on medical AI with its Baichuan-M reasoning models.
- Command R — Cohere's open-weight LLM family optimized for retrieval-augmented generation and tool use; succeeded by Command A as Cohere's flagship.
- DBRX — Open-weight mixture-of-experts LLM released by Databricks in 2024; retired from Databricks serving in 2025, though the weights remain available.
- DeepSeek — Open-weight LLM family and chat assistant from Hangzhou-based DeepSeek, known for strong reasoning models. The official DeepSeek-V4-Flash-0731 API brings a major agent-capability upgrade via post-training, with native Responses API support and Codex adaptation.
- Falcon — Family of open-weight large language models developed by the Technology Innovation Institute (TII).
- Gemma — Google's family of lightweight open-weight models derived from the same research as Gemini.
Sources
This record was last reviewed on 2026-07-30.
Machine-readable record: /api/products/phi.json · Spot an error? Suggest a correction