Phi-4
Microsoft's small language models - Phi-4 leads SLM benchmarks
Microsoft · Language Models
Quick Answer
Phi-4 is microsoft's small language models - phi-4 leads slm benchmarks, made by Microsoft. It is free and open source. Key uses: On-device AI, Mobile apps, Low-latency inference.
Last updated:
What is Phi-4?
Phi-4 (Dec 2024) is Microsoft's latest small language model at 14B parameters, surpassing GPT-4o mini and Gemini 1.5 Pro on STEM and reasoning benchmarks. Phi-4-mini (3.8B) and Phi-4-multimodal add vision and audio capabilities. Phi-4-reasoning (May 2025) adds chain-of-thought.
What's new in Phi-4 in 2026?
Phi-4-reasoning (May 2025) - #1 open model on AIME math; Phi-4-mini-reasoning for edge devices; Phi-4-multimodal adds speech & image understanding; all released under MIT license
What can you do with Phi-4?
- On-device AI
- Mobile apps
- Low-latency inference
- Cost-efficient deployment
- STEM reasoning
Pros
- Phi-4 14B beats much larger models on STEM
- Open weights (MIT license)
- Multimodal support
- Reasoning model available
- Runs on consumer hardware
Cons
- Limited for very long documents
- Less well-known outside enterprise
What are the best alternatives to Phi-4?
Compare Phi-4 Head-to-Head
Frequently Asked Questions about Phi-4
Is Phi-4 free?
Yes, Phi-4 is fully free and open source.
What is Phi-4 used for?
Phi-4 is used for: On-device AI, Mobile apps, Low-latency inference, Cost-efficient deployment, STEM reasoning.
What are the best alternatives to Phi-4?
Top alternatives to Phi-4 include Mistral AI, Llama, Gemma.