# Phi-4

Microsoft's compact 14B dense reasoning model, MIT-licensed, tops MMLU in its size class with 16K context.

## Agent Decision Summary
- Risk level: low
- Source confidence: low
- Recommended workflows: Local or private AI stack
- Permission surface: memory
- Agent JSON: https://www.openagent.bot/models/phi-4.agent.json

## Summary
Microsoft's compact 14B dense reasoning model, MIT-licensed, tops MMLU in its size class with 16K context.

## What It Does
Phi-4 is a model in the models category. Microsoft's compact 14B dense reasoning model, MIT-licensed, tops MMLU in its size class with 16K context.

## How To Evaluate
Evaluate Phi-4 by starting from the official sources, checking its docs, demo interface surface, and running one narrow workflow before expanding scope. Recorded integrations include GitHub, Hugging Face.

## Why It Matters
Phi-4 delivers state-of-the-art reasoning performance in the 14B parameter class with a permissive MIT license, making it the go-to compact model for developers who need strong reasoning on modest hardware.


## Best For
- Developers seeking a compact, MIT-licensed reasoning model for local deployment
- Teams that need top MMLU performance in the sub-20B parameter range
- Builders looking for a lightweight open model that runs efficiently on consumer hardware

## Not For
- Use cases requiring very long context beyond 16K tokens
- Workloads that need a large MoE architecture with hundreds of billions of parameters

## Fit Matrix
- Local or private AI stack: strong. Phi-4 has multiple signals for local or private ai stack, including matching tags, capabilities, category, or positioning. Required check: Verify hardware requirements, data path, storage, and whether all calls stay in your environment.
- Browser automation: partial. Phi-4 has at least one signal for browser automation, but should be checked against a real task before adoption. Required check: Run one non-sensitive website task and inspect clicks, waits, retries, and changed URLs.
- Memory or RAG workflow: partial. Phi-4 has at least one signal for memory or rag workflow, but should be checked against a real task before adoption. Required check: Create, update, retrieve, correct, and delete memory or retrieval objects with real data.
- Coding agent workflow: weak. Phi-4 is not primarily positioned for coding agent workflow in the current metadata. Required check: Run a small repository change and inspect the diff, tests, and rollback path.
- Connector or protocol layer: weak. Phi-4 is not primarily positioned for connector or protocol layer in the current metadata. Required check: Connect one low-risk service, then inspect schemas, auth scope, errors, and logs.
- Evaluation and observability: weak. Phi-4 is not primarily positioned for evaluation and observability in the current metadata. Required check: Add one repeatable test case and confirm results can run again in review or CI.

## Evidence
- verified: Phi-4 is listed as open source. Source: License metadata: MIT
- inferred: Phi-4 supports these recorded deployment modes: local, self hosted, cloud. Source: OpenAgent decision signal metadata.
- inferred: Phi-4 is tagged with local inference, inference capabilities. Source: OpenAgent capability taxonomy.

## Missing Checks
- GitHub repository has not been recorded.
- Dedicated docs link is missing.
- Repository freshness has not been recorded.

## Next Actions
- Open Hugging Face: https://huggingface.co/microsoft/phi-4

## Facts
- Category: models
- Resource type: model
- Open source: yes
- License: MIT
- Last verified: 2026-06-16
- GitHub stars: 12000

## Capabilities
- local-inference
- inference

## Structured Use Case Tags
- local-ai
- self-hosted-ai

## Links
- Hugging Face: https://huggingface.co/microsoft/phi-4

## Structured Outputs
- JSON: https://www.openagent.bot/models/phi-4.json
- Markdown: https://www.openagent.bot/models/phi-4.md
- Agent JSON: https://www.openagent.bot/models/phi-4.agent.json
- Canonical: https://www.openagent.bot/models/phi-4