Skip to content

Phi 4

Microsoft
Code

Phi 4 is a 14.66-billion-parameter dense transformer from Microsoft, trained on 9.8 trillion tokens with an emphasis on curated synthetic data for advanced reasoning. It outperforms many larger models on science and math benchmarks, making it a strong choice for reasoning-intensive workloads at moderate scale. The model focuses on English with capabilities in code generation and mathematical problem solving. A 16K context window and flash attention allow efficient inference, and it quantizes well to GGUF for self-hosted GPU deployments.

Hardware Configuration

Optional — for precise deployment recommendations
Quantization Quality Size Fit
Q8_0 High 14.51 GB
Q6_K High 11.2 GB
Q5_K_S Medium 9.45 GB
Q4_K_S Medium 7.86 GB
Q4_0 Medium 7.81 GB
Q4_1 Medium 8.63 GB
Q3_K_S Low 6.06 GB
Q2_K Low 5.17 GB
Q3_K Low 6.86 GB
Q3_K_L Low 7.39 GB
Q4_K Low 8.43 GB
Q5_0 Low 9.45 GB
Q5_1 Low 10.28 GB
Q5_K Low 9.88 GB
Last updated: March 5, 2026