Microsoft · The Reasoning-Model Era
Phi-4
Microsoft's small-but-capable reasoning model. Punches above its 14B parameter count.
By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026
Phi-4 is the model Microsoft research built to demonstrate that training data quality matters more than dataset size. The 14B parameter model matches or exceeds models two to four times larger on reasoning-heavy benchmarks. It is the canonical example of the 'textbook-quality data' thesis.
Field signature
Handles multi-step math problems better than its size suggests.
Specifications
| Released | 2024-12 |
|---|---|
| Context window | 16,000 tokens |
| Pricing | Free self-hosted |
| Modalities | text |
| License | MIT |
| Era | The Reasoning-Model Era |
Strengths
- Reasoning per parameter
- Small deployment footprint
- Open weights
Weaknesses
- Short context window
- No multimodal
Authentication markers
The fingerprints by which Phi-4 can be identified from its output alone.
| Tell | Meaning |
|---|---|
| Strong math but context window hits quickly. | Phi family. |
Notable works
- Demonstrated textbook-data thesis at scale
Market position
Free self-hosted
Partner offer
Partner offerings listed for operator convenience. See disclosure for terms.
View partner →Affiliate link — see disclosure.
Primary sources
- [1] Microsoft: Phi-4
From the Almanac shop
The Operator's Compendium
Every agent harness, every routing pattern, every cost trick. 90-page PDF.
$29 — Coming soon