Phi-4 Multimodal Instruct
Phi-4 Multimodal Instruct adds image and audio processing to a Phi-4 Mini language backbone. It produces text, so reading a recording and speaking a reply are separate parts of an application.
What distinguishes this release
The card lists 23 text languages, English for vision and eight audio languages. Its image and speech encoders use modality-specific adapters. These distinct language lists matter when selecting a multilingual document or recording workflow.Phi-4 Multimodal Instruct official model card
Build the task around its input and output
Practical guidance. A meeting assistant can use audio to produce a transcript or answer, then pass approved text to a separate speech system if a spoken reply is required. Preserve timestamps and source files outside the model so a reviewer can return to the original recording.
Review the result in its intended setting
Practical guidance. For mixed document input, review small print, chart labels and image crops independently. Do not assume that support for a language in text also establishes the same quality for speech or photographed text. Choose examples that reflect the actual input channel.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | microsoft/Phi-4-multimodal-instruct |
| Context window | 131,072 tokens Published checkpoint context; serving engines may impose lower limits Phi-4 Multimodal Instruct official model card |
| Model size | 5.6B multimodalPhi-4 Multimodal Instruct official model card |
| Input and output | Text, images and audio in; text outPhi-4 Multimodal Instruct official model card |
| Weight license | MITPhi-4 Multimodal Instruct official model card |