Models/Phi-4-mini-instruct

MDL

Phi-4-mini-instruct

microsoft

Hugging Face ↗

Verified against Artificial Analysis · Jul 13, 2026

Capability

percentile index

Price /M

$0 / $0

in / out

Context

131.1K

tokens

Avg score

24.2

15 benchmarks

AA Intelligence

6.0

index

Speed

46 tok/s

TTFT 0.3s

Released

Feb 26, 2024

About

Phi-4-mini-instruct is a open-weights model in the Phi family from microsoft. Benchmarked on 15 evals, averaging 24.2.

mittext-generation

Benchmark scores · 15

avg 24.2 — higher is better

IFEvalOpen LLM LB · 2026-05

73.8

MATHartificial-analysis · 2026-07

69.6

MMLU-Proartificial-analysis · 2026-07

46.5

BIG-Bench HardOpen LLM LB · 2026-05

38.7

GPQA Diamondartificial-analysis · 2026-07

33.1

IFBenchartificial-analysis · 2026-07

21.1

MATH Lvl 5Open LLM LB · 2026-05

17.0

AA Long-Context Reasoningartificial-analysis · 2026-07

13.7

LiveCodeBenchartificial-analysis · 2026-07

12.6

SciCodeartificial-analysis · 2026-07

10.8

τ²-benchartificial-analysis · 2026-07

8.2

AIMEartificial-analysis · 2026-07

6.7

MuSROpen LLM LB · 2026-05

6.4

Humanity's Last Examartificial-analysis · 2026-07

4.2

Terminal-Bench Hardartificial-analysis · 2026-07

The research behind this model

all papers mentioning it →

Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs

Details

Publishermicrosoft

FamilyPhi

Parameters—

Licensemit

WeightsOpen

ReleasedFeb 26, 2024

Performance & price

Output speed46 tok/s

Latency (TTFT)0.3s

Price in / out$0 / $0 per 1M

Coding index3.8

Math index6.7

Median across API providers, via Artificial Analysis. Blended = 3:1 input:output. For reasoning models, latency includes thinking time. Methodology →

History

full ledger →

Capability over time

Append-only ledger — every observed change to this model's numbers.

API providers

Artificial Analysis$0/M