Model · Microsoft · Text
Phi-4
Microsoft's small model trained heavily on synthetic data. Punches above its 14B weight on reasoning and math; MIT-licensed and runs locally.
- Modality
- Text
- License
- MIT (Open)
- Parameter size
- 14B
- Context window
- 16,384 tokens
- Released
- December 12, 2024
- Last verified
- October 28, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 16 GB (24 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 12 GB (16 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Runs on a 16 GB Mac in Ollama, LM Studio or MLX; it only handles a short context (about 16K tokens), so memory use stays low.
Numbers are for: 4-bit build of the 14B (about 9 GB). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- MIT license — fully open
- Strong reasoning and math for a 14B model
- Runs on a single 24GB GPU
Weaknesses
- 16k context is small
- Less general-purpose than larger open models
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | MIT |
| Ollama | local | ollama run phi4 |
Official sources
- Model card model
- Technical report paper
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama phi4 page (9.1 GB); mlx-community/phi-4-4bit.
- — Not deprecated: Microsoft has released no Phi-5, only Phi-4 variants (reasoning, mini, multimodal, and Phi-4-reasoning-vision-15B in January 2026). For local use it is now outclassed by Gemma 4 E4B / 12B and Qwen3.5 9B, which offer 128K-262K context against Phi-4's 16K, plus image input.
- — Verified current against the model card — 14B, 16K context, MIT, no deprecation notice. Phi-5 does not exist.
- — Initial entry.
Esc