Dev, web & IT
Design, code, host and ship sites and internal tools, with an AI editor doing most of the typing.
Independent reviews. IonBin has no partnerships with the tools on this page and was not paid for these listings. Our evaluations are our own, based on public documentation and our research. Product names are trademarks of their owners and are used for identification only. How disclosure works.
Model · Microsoft
Stack 08
Phi models (Microsoft)
Verified Oct 17, 2026 Review by Nov 16, 2026 5 sources
Short answer
Phi is Microsoft's family of small language models, released under the MIT License. The newest on Hugging Face include Phi-4-reasoning-vision-15B, a 15B text-and-image model. Phi models run on Azure AI Foundry, Hugging Face and Ollama, and can be fine-tuned. They suit on-device or low-cost tasks where a frontier model is unnecessary.
What it is
Phi is a family of small language models from Microsoft. Microsoft describes them as open models under the MIT License, available through Azure and common open-model hubs. The Azure product page names Phi-4-mini and Phi-4-multimodal. Microsoft’s Hugging Face page shows newer releases, such as Phi-4-reasoning-vision-15B, released March 4, 2026.
How it works
- Download a Phi model from Hugging Face or pull it with Ollama to run locally, or use Azure AI Foundry with pay-as-you-go pricing.
- Fine-tune it on your own examples for a narrow task such as classification or extraction.
- Use it where latency, cost or offline operation matter more than broad knowledge.
Specs and pricing (checked 2026-10-17). Phi-4-mini-instruct supports a 128K-token context. Phi-4-reasoning-vision-15B has 15 billion parameters, takes text and images, and has a 16,384-token context. The weights carry no license fee under MIT. Azure’s hosted rates aren’t listed on the product page, so check Azure’s pricing before you budget.
Benefits
- Permissive MIT License for commercial use.
- Small enough to run on modest hardware (see running models locally).
- A good starting point for teaching a model one narrow, repeatable task.
Limits and risks
- Less capable. Small models know less and reason less than frontier models. Test on your own tasks.
- Short context on some models. Context windows vary widely, from 16K tokens to 128K, so check the model card.
- Out-of-date knowledge. The Azure page lists a June 2024 training cutoff for the models it describes, so recent facts are missing.
Checked specs
Snapshot from official sources on Oct 17, 2026. Model lineups and prices change often; this page is re-verified every 30 days.
| Type | Small language models, MIT License [1] |
| Recent release | Phi-4-reasoning-vision-15B: released March 4, 2026; 15B parameters; text and images in, text out; 16,384-token context; MIT [2] |
| Phi-4-mini-instruct | 128K-token context; MIT License [3] |
| Where to run | Azure AI Foundry (incl. pay-as-you-go), Hugging Face, Ollama [1] |
| Pricing (checked 2026-10-17) | No license fee for the weights (MIT); hosted use on Azure is pay-as-you-go, and the product page doesn't list rates [1] |
| Training cutoff (as listed on the Azure page) | June 2024 [1] |
| Customization | Can be fine-tuned [1] |
Best-fit stacks
Agent compatibility
| API access | Yes | Official docs ↗ |
| Also offered via | Azure AI Foundry; Hugging Face; Ollama | |
Levels come from each tool's recorded, sourced compatibility. How the levels work.
Related entries
Sources
- Phi open model family · Microsoftofficial accessed Oct 17, 2026
- Phi-4-reasoning-vision-15B model card · Microsoftofficial accessed Oct 17, 2026
- Phi-4-mini-instruct model card · Microsoftofficial accessed Oct 17, 2026
- Model Hub documentation · Hugging Faceofficial accessed Oct 17, 2026
- Ollama · Ollamaofficial accessed Oct 17, 2026
Practitioner views are restated in our words and labeled as views, not measured results. Product facts come from official sources dated above.
Suggest an edit
IonBin's wiki isn't user-edited. Corrections go to our Directory Curator, who checks them against primary sources before anything changes. Spot something wrong or out of date on “Phi models (Microsoft)”? Email us with a link to a primary source.
Suggest an edit by email or write to [email protected]