NEWHow AI models draw the world map · Gecko Tests
Beta
Home/Models/Phi 4 Mini Instruct
Microsoft logo

Phi 4 Mini Instruct

by Microsoft · Released Oct 2025

Open Source
48.4
avg score
Rank #156
Compare
Better than 50% of all models
Context
131K tokens (~66 books)
Input $/1M
$0.08
Output $/1M
$0.35
Type
text
License
Open Source
Benchmarks
7 tested
Data as of
About

Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data. The model belongs to the Phi-4...

Tested on 7 benchmarks · BenchGecko score 48.4. Top scores: IFEval (73.8%), BBH (HuggingFace) (38.7%), MMLU-PRO (32.6%).

Looking for similar performance at lower cost?
Gemma 4 26B A4B scores 47.8 (99% as good) at $0.07/1M input · 16% cheaper
Capabilities
reasoning
6.5
#199 globally
math
17.0
#243 globally
knowledge
20.3
#270 globally
general
38.7
#73 globally
language
73.8
#72 globally
speed
3.0
#130 globally
Benchmark Scores
Compare All
Tested on 7 benchmarks · Ranked across 6 categories
Score Distribution (all 312 models)
0255075100
▲ You are here
reasoningCompare reasoning →
MUSR

HuggingFace MuSR (Multi-Step Reasoning). Tests multi-hop reasoning requiring chaining multiple facts together.

6.5·
mathCompare math →
MATH Level 5

HuggingFace evaluation of MATH Level 5 problems. Competition math requiring advanced reasoning and proof construction.

17.0·
knowledgeCompare knowledge →
MMLU-PRO

HuggingFace MMLU-Pro. Harder version of MMLU with 10 answer choices instead of 4 and more challenging questions.

32.6·
GPQA

HuggingFace evaluation of GPQA (Graduate-Level Google-Proof Q&A). PhD-level science questions that cannot be easily searched.

7.9·
Quick compare:Alibaba logovs Qwen3 4B Thinking 2507Microsoft logovs Phi 4 Mini InstructOpenAI logovs GPT-5 Pro
Excellent (85+) Good (70-85) Average (50-70) Below (<50)
Model Family · Microsoft Phi 4
Jan 2025
49.0
$0.07/M in16Kctx24 benchmarks
Oct 2025
48.4-0.6
$0.08/M in(+0.01)131Kctx(+115K)7 benchmarks
Feb 2025
48.4
n/an/actx17 benchmarks
Feb 2025
n/an/actx4 benchmarks
See the full Phi 4 family →
Similar Models
Alibaba logo
Qwen3 4B Thinking 2507
Alibaba
48.4n/a
Microsoft logo
Phi 4 Mini Instruct
Microsoft
48.4n/a
OpenAI logo
GPT-5 Pro
OpenAI
48.7$15.00/1M
Links
Info
MicrosoftPricing explorerDevelopers · API
Research
Technical Report
Documentation
API DocsPlayground
Community
@Microsoft
Source Code
GitHubHugging Face
BenchGecko API
phi-4-mini-instruct
Specifications
  • Typetext
  • Context131K tokens (~66 books)
  • ReleasedOct 2025
  • LicenseOpen Source
  • StatusActive
  • Cost / Message~$0.001
Available On
Microsoft logoMicrosoft$0.08
Categories
reasoningmathknowledgegenerallanguagespeed
Learn More
context-windowtransformeropen weightstokens
Share & Export
Tweet
Related Models
Alibaba logoQwen3 4B Thinking 2507
48.4
Microsoft logoPhi 4 Mini Instruct
48.4
OpenAI logoGPT-5 Pro
48.7
Meta logoLlama 3.3 70B Instruct
47.9
Phi 4 Mini Instruct is an open-source text AI model by Microsoft, released in October 2025. It has an average benchmark score of 48.4. Context window: 131K tokens.

Related Models

Qwen3 4B Thinking 2507 · AlibabaPhi 4 Mini Instruct · MicrosoftGPT-5 Pro · OpenAILlama 3.3 70B Instruct · MetaGLM 5.3 Flash · z-ai

Benchmarks

IFEvalBBH (HuggingFace)MMLU-PROMATH Level 5GPQA

Related Pages

Microsoft · ProviderMicrosoft · EconomyAll ModelsCompare ModelsPricingDevelopers · APIContext Window · Glossary

Key facts · as of 2026-06-22

  • Phi 4 Mini Instruct by Microsoft. BenchGecko score 48.4, rank 156 of 312 scored models (normalized average of public benchmark scores).
  • List price $0.0800 input · $0.35 output per 1M tokens (as of 2026-06-22).

How to cite · data as of 2026-06-22

Phi 4 Mini Instruct · benchmarks, pricing and providers. BenchGecko, data as of 2026-06-22. https://benchgecko.ai/model/phi-4-mini-instruct

Credit "Source: BenchGecko" with a link. Prices per provider and Gecko Tests are BenchGecko data (CC BY 4.0); benchmark scores keep their original source, listed in the JSON. JSON · llms.txt · MCP