Model

Phi-1.5 (1.3B)

OPEN

Microsoft · Phi-1.5 family · released Sep 11, 2023

Base LM extending Phi-1's textbook-quality synthetic data to common-sense reasoning and NLU; near-SOTA under 10B params, MIT-licensed.

ReasoningCodingVisionFunction callingTool useAgentic

-148.3

Elo · rank #417

Parameters

1.3B

Active params

1.3B (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

ARC-ChallengeReasoning44.4%#138

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning75.6%#35

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning7.5%#142

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning2.4%#290

best: GPT-5.6 · 94.6%

GSM8KMath40.2%#144

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning47.6%#151

best: Claude 3 Opus · 95.4%

HumanEvalCoding34.1%#144

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning20.3%#154

best: Gemma 4 26B A4B · 98.5%

MATH-500Math1.8%#205

best: GPT-5 · 99.4%

MBPPCoding37.7%#89

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

MMLU-ProKnowledge7.7%#184

best: Claude Fable 5 · 91.5%

MMLUKnowledge37.6%#260

best: OpenAI o3 · 92.9%

OpenBookQAReasoning37.2%#47

best: Claude 1 · 90.8%

PIQAReasoning76.6%#67

best: GPT-4o mini · 93.1%

Social IQaReasoning52.6%#18

best: Apple DCLM-Baseline 7B · 82.9%

WinoGrandeReasoning73.4%#78

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

1 GB

VRAM @ FP16

2.6 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

Phi-1.5 family

Elo progression across releases

Phi-1.5 (1.3B)Sep 2023-148.3

Resources

HFmicrosoft/phi-1_5

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