Phi-4
Phi-4 is an open-weight language model from Microsoft, released on December 12, 2024.
- Developer
- Microsoft
- Released
- December 12, 2024
- Total parameters
- 14B
- Architecture
- Dense
- Context window
- 16K
- Input
- Text
- License
- MIT
- Commercial use
- Yes
- Official weights, GB
- 29.3 GB
- Hugging Face repository
- huggingface.co
- Checked on
Phi-4 is a dense model with 14B parameters. Microsoft lists a context window of 16K tokens. The model takes text input. The weights are released under the MIT license, which allows commercial use. The official weights on Hugging Face take about 29.3 GB in BF16. Running the model needs at least that much memory across GPUs and system RAM, plus room for the context cache; quantized versions need less.
- Total parameters, billions
- 14B
- Context window, tokens
- 16K
- License type
- MIT
- License conditions
- MIT: keep the copyright and permission notice.
- Weights precision
- BF16
- Official quantized or alternative versions
- microsoft/phi-4-gguf; microsoft/phi-4-onnx
- License text
- huggingface.co
- Sources
Page URL Hugging Face model card https://huggingface.co/microsoft/phi-4/raw/main/README.md Hugging Face API https://huggingface.co/api/models/microsoft/phi-4?expand=safetensors
Questions
What license does Phi-4 use?
The weights are released under the MIT license, which allows commercial use; keep the copyright and permission notice.
How much memory does Phi-4 need?
The official BF16 weights take about 29.3 GB. Microsoft also publishes GGUF and ONNX versions (microsoft/phi-4-gguf and microsoft/phi-4-onnx); the memory calculator checks which setups fit.
How many parameters does Phi-4 have?
Phi-4 is a dense model with 14B parameters.
What context window does Phi-4 have?
Microsoft lists a context window of 16K tokens.
Sources
Every figure comes from the developer’s model card, license text and official announcements, and from the model’s Hugging Face repository metadata. Each model page lists its sources and the date they were checked; the register is rechecked every week.