Loading page
LightweightMicrosoft
Phi-4
Phi-4 is Microsoft's lightweight language model designed for efficient text generation with a 16K token context window.
Context 16K
Tier Lightweight
Input from
$0.070 / 1M tokens
across 2 providers
API Pricing
| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $0.070 | $0.140 | |
| $0.070 | $0.140 |
Prices updated daily. Last check: Sep 8, 2026
Phi-4 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
1.0 / 100
Math
18.0 / 100
Output Speed
43.8 t/s
Latency (TTFT)
1.0s
Reasoning & Knowledge
- MMLU-Pro71.4%
- GPQA Diamond57.5%
- Humanity's Last Exam3.8%
Coding
- LiveCodeBench23.1%
Math
- AIME 202518.0%
- AIME14.3%
- MATH-50081.0%
Agentic & Tool Use
- Terminal-Bench Hard3.8%
- τ²-bench0.0%
Instruction & Long Context
- IFBench23.5%
- Long-Context Reasoning0.0%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Microsoft
- Family
- Phi
- Tier
- Lightweight
- Context Window
- 16K
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Fast inference speed at 17.24 output tokens per second
- Quick response initiation with 354ms time to first token
- Lightweight architecture for efficient deployment
- 16K token context window supports moderate-length conversations
- Part of Microsoft's established Phi model family
- Optimized for text generation tasks
Limitations
- No tool calling or function execution support
- Text-only modality - no image or multimodal input
- Proprietary model with no open source access
- Smaller context window than flagship models
- Limited to lightweight tier capabilities
Key Features
•16,384 token context window
•Text input and output processing
•Streaming response generation
•Microsoft Azure integration
•Lightweight model architecture
•Fast inference optimization
About Phi-4
Phi-4 is Microsoft's lightweight language model in the Phi family, positioned as an efficient option for text-based applications. As a lightweight tier model, it offers a balance between capability and computational efficiency compared to larger flagship models in Microsoft's lineup.
The model supports text-only input and output with a 16,384 token context window. Phi-4 delivers 17.24 output tokens per second with a time to first token of 354 milliseconds according to Artificial Analysis benchmarks. The model does not include tool calling capabilities and is available as a proprietary offering rather than open source.
Phi-4 serves applications where computational efficiency is prioritized alongside reasonable language understanding capabilities. It competes with other lightweight models in scenarios requiring faster inference speeds and lower resource consumption than flagship alternatives.
Common Use Cases
Phi-4 is suited for applications requiring efficient text processing where speed and resource efficiency are priorities over maximum capability. This includes chatbots with moderate complexity requirements, content generation for blogs or marketing copy, text summarization of shorter documents, and educational applications where quick responses enhance user experience. The model's lightweight nature makes it appropriate for scenarios with high-volume requests or resource-constrained environments where deploying larger flagship models would be impractical or costly.
Frequently Asked Questions
How much does Phi-4 cost per million tokens?
Phi-4 pricing varies by provider and pricing type. Check the pricing table above for current rates across all providers offering this model.
What is Phi-4 best used for?
Phi-4 excels at efficient text generation tasks including chatbots, content creation, and text summarization where fast response times and resource efficiency are important. Its lightweight design makes it suitable for high-volume applications or environments with computational constraints.
Does Phi-4 support tool calling or multimodal input?
No, Phi-4 is a text-only model that does not support tool calling, function execution, or multimodal inputs like images. It is focused on efficient text processing and generation tasks.