Pyyan / Open-Weight Models / Phi-4

Phi-4

Best small model — fits a 12GB GPU.

currentMicrosoft ↗Verified 2026-08-13

84.8%

MMLU

Specifications

Licence	MIT
Parameters	14B

Official links

Microsoft official site

What matters

Beats GPT-4o on mathematics
Phi-4-mini runs in ~3GB at Q4

Best for local use.

Compare with

All 5 side by side →

Phi-4 vs DeepSeek V4-Pro →Phi-4 vs Kimi K3 →Phi-4 vs GLM-5.2 →Phi-4 vs Qwen3-235B-A22B →Phi-4 vs Llama 5 →

Also in Open-Weight Models

DeepSeek V4-Pro · DeepSeek Kimi K3 · Moonshot AI GLM-5.2 · Zhipu AI Qwen3-235B-A22B · Alibaba Llama 5 · Meta