Phi-4 Mini

Training compute

10×10²² FLOP

Parameters

3.8B

Published

Mar 3, 2025

Phi-4 Mini is an AI model developed by Microsoft (United States), first published in March 2025. It works in the language domain, on tasks such as language modeling/generation, visual question answering, code generation and 2 more.

Training it took an estimated 10×10²² FLOP of compute (estimation method: operation counting,hardware). The model has 3,800,000,000 parameters. It was trained on roughly 5T datapoints. Training ran on 512 NVIDIA A100 SXM4 80 GB for about 504 hours.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Language
Task: Language modeling/generation, Visual question answering, Code generation, Quantitative reasoning, Translation
Training compute: 10×10²² FLOP
Compute estimation method: Operation counting, Hardware
Parameters: 3,800,000,000
Dataset size: 5T
Training hardware: NVIDIA A100 SXM4 80 GB
Chips used: 512
Training time: 504 h
Training power draw: 402.0 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident