Phi-4-Multimodal

Training compute

1.2×10²³ FLOP

Parameters

5.6B

Published

Mar 3, 2025

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

Training it took an estimated 1.2×10²³ FLOP of compute (estimation method: operation counting,hardware). The model has 5,600,000,000 parameters. Training ran on 512 NVIDIA A100 SXM4 80 GB for about 672 hours.

Access: Open weights (unrestricted). Its weights are openly available. It is built on top of Phi-4 Mini. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Multimodal, Language, Vision, Speech
Task: Language modeling/generation, Question answering, Visual question answering, Speech recognition (ASR), Translation, Audio question answering, Character recognition (OCR)
Training compute: 1.2×10²³ FLOP
Compute estimation method: Operation counting, Hardware
Parameters: 5,600,000,000
Training hardware: NVIDIA A100 SXM4 80 GB
Chips used: 512
Training time: 672 h
Training power draw: 402.0 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Base model: Phi-4 Mini
Epoch confidence: Confident