Phi-4 / Phi family is a SLM from Microsoft (United States).
One-line fit: best used for Edge, privacy, cheap routing.
Key specifications
| Provider | Microsoft |
|---|---|
| Model | Phi-4 / Phi family |
| Type | SLM |
| Context window | Varies |
| Pricing (indicative) | Low / on-device |
| Company category | Hyperscaler / Models |
| Region | United States |
| Official site | https://www.microsoft.com/ai |
Benchmarks and public standing
Best-in-class small-model reasoning for size
Treat scores as directional. SWE-bench, Terminal-Bench, LMSYS Arena, Artificial Analysis, and vendor cards use different harnesses and are not always comparable 1:1. Re-check the latest official model card before decisions.
What this model is good at
Edge, privacy, cheap routing
This aligns with Microsoft’s broader strengths:
- Enterprise Copilot
- Azure compliance regions
- GitHub Copilot UX
- On-device Phi models
Limitations and watch-outs
Not frontier long-horizon agents
- Rate limits, regional availability, and data retention policies vary by plan.
- Agent harness quality (Cursor, Claude Code, Codex, custom tools) can change outcomes more than raw model Elo.
- Open-weight availability (if any) is separate from hosted API quality and safety filters.
Ideal users
- Product engineers shipping features that match: Edge, privacy, cheap routing
- Teams standardizing on the Microsoft ecosystem
- Agent builders who need a SLM
When to pick something else
- Cheaper volume: compare lower tiers from the same lab or open Chinese/EU alternatives.
- Maximum hard SWE thoughtfulness: compare Claude Fable-class and other coding flagships.
- Giant multimodal corpora / Workspace: compare Gemini-class models.
- Self-host / open weights: Llama, Qwen, DeepSeek, GLM, Mistral open lines.
Parent company
Microsoft – Azure OpenAI partner at scale, Copilot across Office/GitHub/Windows, and first-party Phi small models.
Related models from Microsoft
- Azure OpenAI hosted frontier – Hosted frontier
- GitHub Copilot models – IDE coding
Data is curated for AIForumSphere model directory mid-2026 directional figures.
