Phi-4
Open Weights
Microsoft
Microsoft's 14.7B-parameter small language model (December 2024), open-sourced on Hugging Face under the MIT license. Trained heavily on synthetic data, Phi-4 matches models 5x its size on math and reasoning, outperforming GPT-4o by ~6 points on GPQA and math while running on consumer hardware.
Strengths
- Matches models 5x its size on math and reasoning benchmarks
- Outperforms GPT-4o by ~6 points on GPQA and math
- 82.6 on HumanEval code generation
- Open weights under permissive MIT license
- Runs on consumer hardware (~12GB GPU)
Caveats
- 16K context smaller than larger frontier models
- Text-only (no vision or audio input)
- 14B capacity limits breadth versus frontier LLMs
- Vendor
- Microsoft
- Context Window
- 16,384 tokens
- License
- MIT
- Release Date
- 2024-12-12
- Modalities
- text
| Benchmark | Score | Max |
|---|---|---|
| HumanEval | 82.6 | 100.0 |
| livebench_instruction_following | 54.8 | 100.0 |
| livebench_language | 29.3 | 100.0 |
| livebench_coding | 30.7 | 100.0 |
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under MIT license - free to download and self-host; runs comfortably on consumer GPUs (~12GB).