Model

Phi-3-mini (3.8B)

OPEN

Microsoft · Phi-3 family · released Apr 23, 2024

First production-grade Phi; ~GPT-3.5-class quality at 3.8B, runs on-device, shipped in a 4K-context variant (this record) and a separate 128K-context sibling, both MIT-licensed.

ReasoningCodingVisionFunction callingTool useAgentic

479.2

Elo · rank #312

Parameters

3.8B

Active params

3.8B (dense)

Context

4K tokens

Architecture

Dense decoder-only Transformer

License

MIT

Languages

—

API price (in/out)

$0.13 / $0.52

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AGIEvalReasoning39.0%#33

best: OLMo 3-Think 32B · 88.2%

ANLIReasoning53.6%#3

best: Phi-3-small (7B) · 58.1%

ARC-ChallengeReasoning86.3%#29

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning73.5%#46

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning30.6%#258

best: GPT-5.6 · 94.6%

GSM8KMath85.7%#64

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning75.3%#111

best: Claude 3 Opus · 95.4%

HumanEvalCoding57.3%#109

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning56.1%#126

best: Gemma 4 26B A4B · 98.5%

MATH-500Math14.2%#187

best: GPT-5 · 99.4%

MBPPCoding69.8%#40

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLU (EU-21 languages)Knowledge44.9%#10

best: Llama 3.1 70B · 77.1%

MMLU-ProKnowledge40.2%#149

best: Claude Fable 5 · 91.5%

MMLUKnowledge70.9%#140

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.38#33

best: Hunyuan-Large (A52B) · 9.4

OpenBookQAReasoning82.2%#7

best: Claude 1 · 90.8%

PIQAReasoning83.5%#15

best: GPT-4o mini · 93.1%

Social IQaReasoning77.6%#7

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge61.4%#34

best: Sarvam-1 (2B) · 90.6%

TruthfulQAKnowledge65.0%#16

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning71.6%#91

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

2.3 GB

VRAM @ FP16

10 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF · MLX · ONNX

Phi-3 family

Elo progression across releases

Phi-3-mini (3.8B)Apr 2024479.2 Phi-3-medium (14B)May 2024757.6 Phi-3-small (7B)May 2024658.0

Resources

HFmicrosoft/Phi-3-mini-4k-instruct DOCSlearn.microsoft.com/en-us/azure/ai-studio/how-to/deploy-models-phi-3 APIAvailable via Microsoft API

API price $0.13/$0.52 · each benchmark row carries its own source badge (see methodology)