Model

Phi-2 (2.7B)

OPEN

Microsoft · Phi-2 family · released Dec 12, 2023

Breakout 2.7B base model matching/beating 7B-13B peers on reasoning, math and code; MIT-licensed, widely adopted for on-device use.

ReasoningCodingVisionFunction callingTool useAgentic

194.4

Elo · rank #364

Parameters

2.7B

Active params

2.7B (dense)

Context

2K tokens

Architecture

Dense decoder-only Transformer

License

MIT

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning29.8%#36

best: OLMo 3-Think 32B · 88.2%

ANLIReasoning42.5%#6

best: Phi-3-small (7B) · 58.1%

ARC-ChallengeReasoning75.9%#42

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning88.5%#9

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning59.4%#80

best: ERNIE 4.5 300B-A47B · 94.3%

GSM8KMath61.1%#114

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning53.6%#146

best: Claude 3 Opus · 95.4%

HumanEvalCoding59.0%#103

best: Claude Opus 4.5 · 99.4%

MBPPCoding60.6%#59

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

MMLUKnowledge56.3%#217

best: OpenAI o3 · 92.9%

OpenBookQAReasoning73.6%#9

best: Claude 1 · 90.8%

PIQAReasoning60.2%#79

best: GPT-4o mini · 93.1%

Social IQaReasoning68.3%#12

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge45.2%#44

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning54.7%#137

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

1.7 GB

VRAM @ FP16

5.4 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 · GPTQ · MLX

Phi-2 family

Elo progression across releases

Phi-2 (2.7B)Dec 2023194.4

Resources

HFmicrosoft/phi-2

API price weights · each benchmark row carries its own source badge (see methodology)