Phi

Large language models (LLMs) software

Generative AI software

Features
Ease of use
Ease of management
Quality of support
Affordability
Market presence

Take the quiz to check if Phi and its alternatives fit your requirements.

Get started

Pricing from

Pay-as-you-go

Free Trial

Free version

User corporate size

Small

Medium

Large

User industry

What is Phi

Phi is a family of small language models designed for text generation and related natural language tasks. It is typically used by developers and researchers who want to run LLMs with lower compute and memory requirements for prototyping, on-device scenarios, or cost-sensitive deployments. The models are published with technical documentation and weights for local or cloud inference, and are commonly integrated via standard ML tooling rather than a proprietary end-user application.

Small-model efficiency focus

Phi models are designed to be relatively small compared with many general-purpose frontier models, which can reduce inference cost and hardware requirements. This makes them practical for experimentation, edge or local deployments, and higher-throughput batch workloads. The smaller footprint can also simplify operationalization where GPU availability is limited.

Developer-friendly distribution

Phi model weights and documentation are commonly distributed in formats compatible with mainstream ML ecosystems. This supports local evaluation, fine-tuning workflows (where permitted by the license), and integration into custom applications. Teams can incorporate the models into existing pipelines without relying on a single hosted API.

Strong baseline for prototyping

As a compact LLM family, Phi can serve as a baseline model for prompt engineering, retrieval-augmented generation (RAG) experiments, and task-specific adaptation. It enables faster iteration cycles due to shorter inference times. This can be useful when comparing multiple model sizes before committing to larger models for production.

Capability ceiling vs larger LLMs

Small models generally underperform larger LLMs on complex reasoning, long-context tasks, and broad knowledge coverage. For high-stakes customer-facing generation, teams may need additional guardrails, retrieval, or escalation to larger models. Performance can vary significantly by task and evaluation setup.

Licensing and usage constraints

Phi releases may include license terms that restrict certain commercial uses or impose conditions on redistribution and modification. Organizations typically need legal review to confirm whether a specific Phi version fits their intended deployment. License terms can also differ across model versions, complicating standardization.

Operational gaps vs hosted platforms

Using Phi as a self-hosted model shifts responsibility for serving, scaling, monitoring, and safety controls to the customer. Compared with fully managed model platforms, teams may need to build or procure additional components for authentication, rate limiting, logging, and content filtering. This increases engineering and MLOps effort for production deployments.

Plan & Pricing

Pricing model: Pay-as-you-go Free tier/trial: Free access available for real-time deployment via Microsoft Foundry and Hugging Face; Azure offers a free account ($200 credit for 30 days) that can be used to experiment with Foundry models. Example costs (selected models — prices are per 1,000 tokens unless noted):

Phi-3-mini (4K or 128K context) — Input: $0.00013 / 1,000 tokens; Output: $0.00052 / 1,000 tokens.
Phi-3.5-mini (128K) — Input: $0.00013; Output: $0.00052.
Phi-3-small (8K or 128K) — Input: $0.00015; Output: $0.00060.
Phi-3-medium (4K or 128K) — Input: $0.00017; Output: $0.00068.
Phi-3.5-MoE (128K) — Input: $0.00016; Output: $0.00064.
Phi-4 (128K) — Input: $0.000125; Output: $0.0005.
Phi-4-mini (128K) — Input: $0.000075; Output: $0.0003.
Phi-4-multimodal (text & image, 128K) — Input: $0.00008; Output: $0.00032.
Phi-4-multimodal (audio, 128K) — Input: $0.004; Output: $0.00032.

Fine-tuning example costs (official figures):

Training: $0.003 per 1,000 tokens (example shown for Phi-3/Phi-4 series).
Hosting (fine-tuned model): $0.80 per hour.
During fine-tuning, Input/Output usage pricing matches the inference input/output rates shown above (e.g., Phi-3-mini input $0.00013 / output $0.00052).

Discount options: Microsoft Foundry pricing page and Azure purchasing options reference volume/commitment and enterprise agreements (contact sales / request quote) — enterprise/provisioned throughput and reservation options are available via Azure sales.

Notes & scope: All figures are taken from Microsoft Azure’s official product and pricing pages and Microsoft announcements (Azure product page, Foundry pricing, and Azure blog announcement). Prices are listed in USD as presented by Microsoft and are charging units per 1,000 tokens for inference; fine-tuning shows training per 1,000 tokens and hosting per hour. Actual charges may vary by region, purchase option, or agreement with Microsoft; see Azure Foundry/pricing for region/currency specifics.

Seller details

Microsoft Corporation

Redmond, Washington, United States

1975

Public

https://www.microsoft.com/

https://x.com/Microsoft

https://www.linkedin.com/company/microsoft/

Tools by Microsoft Corporation

Azure Command-Line Interface (CLI)

Microsoft Azure Red Hat OpenShift

Microsoft Build of OpenJDK

›

Microsoft Visual Studio App Center

Generative AI & LLM	AI code generation software AI image generators software AI video generators AI writing assistants Large language models (LLMs) software
Agents, autonomous & workflow automation	AI chatbots software AI customer support agents software Bot platforms software General-purpose AI agents
Vertical AI	Data science and machine learning platforms Machine learning software
Sales	CPQ software CRM software E-signature software Sales enablement software
Marketing	Email marketing software Marketing automation software SEO tools Social media management tools
Security	Antivirus software Firewall software Identity and access management (IAM) software
Analytics	Analytics platforms Data visualization tools
Collaboration & productivity	Collaborative whiteboard software Video conferencing software
Commerce	E-commerce platforms Payment processing software
Content management	Document management software Knowledge base software Website builder software
Customer service	Customer service automation software Customer success software Help desk software Live chat software
Development	Cloud platform as a service (PaaS) software
ERP	Accounting software ERP systems Expense management software Project management software
HR	Applicant tracking systems (ATS) Payroll software Time tracking software
IT infrastructure	Data warehouse solutions ETL tools Infrastructure as a service (IaaS) providers iPaaS software
IT management	Business process management software Robotic process automation (RPA) software Workflow management software

Phi

What is Phi

Small-model efficiency focus

Developer-friendly distribution

Strong baseline for prototyping

Capability ceiling vs larger LLMs

Licensing and usage constraints

Operational gaps vs hosted platforms

Plan & Pricing

Seller details

Tools by Microsoft Corporation

Popular categories

Generative AI & LLM

Agents, autonomous & workflow automation

Vertical AI

Sales

Marketing

Security

Analytics

Collaboration & productivity

Commerce

Content management

Customer service

Development

ERP

HR

IT infrastructure

IT management