Microsoft (Phi) models

Microsoft Research's Phi family focuses on small language models trained heavily on synthetic and curated data. The current lineup includes Phi-4, Phi-4-mini, the Phi-4-reasoning models and the March 2026 Phi-4-reasoning-vision-15B. Models are published on Hugging Face under MIT, with ONNX and GGUF builds for local use.

2 models trackedData checked October 10, 2026

Lineup

Model	Status	Weights	Tier	Context	Input / 1M	Output / 1M	Released
Phi-4-reasoning-vision-15BPhi-4	Current	Open weights	Compact	16K	—	—	Mar 4, 2026
Phi-4-mini-instructPhi-4	Current	Open weights	Compact	128K	—	—	Feb 2025

The most capable tier in the lineup is compact, led by Phi-4-reasoning-vision-15B. The longest context window is 128K tokens (Phi-4-mini-instruct).

Frequently asked questions

What models does Microsoft (Phi) offer?

As of October 10, 2026, we track 2 Microsoft (Phi) models: Phi-4-reasoning-vision-15B, Phi-4-mini-instruct. Status, prices and context windows for each are in the table on this page.

How do I access Microsoft (Phi) models?

Through Microsoft (Phi)'s developer platform (huggingface.co/microsoft), and for many models through cloud marketplaces and inference providers listed on each model page.

Does Microsoft (Phi) release open-weight models?

Phi models, including Phi-4-mini-instruct and Phi-4-reasoning-vision-15B, are released under the MIT license.

Other providers

OpenAI Anthropic Google xAI Amazon (AWS)Meta Perplexity Mistral AI DeepSeek Alibaba (Qwen)NVIDIA Moonshot AI Z.ai (Zhipu AI)MiniMax IBM Ai2 (Allen Institute for AI)

Still choosing a model?

Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.

Free and vendor-neutral: no model provider pays for placement.
Your details go only to partners you agree to.