Project AURIONAI Use-Cases Research & Innovation
edge LLM

Phi-4-mini-instruct

Microsoft 3.8B small model for edge and mobile, MIT licensed, not gated.

Microsoft AIMicrosoft AIUSAphi-4Family
Microsoft AI
Levelmodel
Classedge LLM
Parameters3.84B
Context128,000
Released19 Feb 2025
Opennessopen weights permissive
Runs onmobile or NPU
Statusactive

Benchmark Results

vendor-reported
GPQAknowledge and reasoning
0-shot CoTvendor-reported
25.2
050100

Score in %. Higher is better.

huggingface.co
GSM8Kmath
vendor-reported
88.6
050100

Score in %. Higher is better.

huggingface.co
MATHmath
vendor-reported
64
050100

Score in %. Higher is better.

huggingface.co
MMLUknowledge and reasoning
5-shotvendor-reported
67.3
050100

Score in %. Higher is better.

huggingface.co
MMLU-Proknowledge and reasoning
vendor-reported
52.8
050100

Score in %. Higher is better.

huggingface.co
Multilingual MMLUknowledge and reasoning
vendor-reported
49.3
050100

Score in %. Higher is better.

huggingface.co
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

successor
Phi-3.5-mini to Phi-4-mini-instruct: successor, medium confidencePhi-3.5-miniPhi-3.5-miniMicrosoft AIPhi-4-mini-instructPhi-4-mini-instructMicrosoft AI, Feb 2025
Microsoft AIPhi-3.5-mini is the parent (successor)medium confidencemodel card
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 2 models. Confidence and evidence are the notes' own.

R&D Topics

Quantization and edge inferenceproduced bySmall language models (SLM)produced by

Identity

Developer
Microsoft AI
Family
phi-4
Released
19 Feb 2025 sourcesregistry date
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
not in the notes
Total parameters
3.84B sources
Active parameters
–
Checkpoint parameters
–
Note
3.8B; safetensors total 3,836,021,760
Context length
128,000 tokens
Input
text
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open weights permissive
Licence
MIT License sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
mobile or NPU
Footprint
Quantized footprint and on-device latency unverified; LiteRT community conversion exists. Trained on 512 A100-80G. 200,064-token vocabulary
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Industry Use Cases

No industry use case links this model.

Sources

  1. huggingface.co/microsoft/Phi-4-mini-instruct https://huggingface.co/microsoft/Phi-4-mini-instruct
  2. huggingface.co/api/models/microsoft/Phi-4-mini-instruct https://huggingface.co/api/models/microsoft/Phi-4-mini-instruct
  3. microsoft / phi-4-mini-flash-reasoning docs.api.nvidia.com https://docs.api.nvidia.com/nim/reference/microsoft-phi-4-mini-flash-reasoning
  4. Phi-4 Mini for Mobile: Microsoft's Small Model on iOS and Android ertas.ai https://www.ertas.ai/blog/phi-4-mini-mobile-ios-android

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Phi-4-Mini-Instruct", https://aurion.semai.id/models/microsoft-phi-4-mini-instruct/, records of October 2026.