Model

GPT-NeoX-20B

OPEN

EleutherAI · GPT-NeoX family · released Feb 9, 2022

Largest publicly available dense LM at its Feb 2022 release; trained on The Pile with the GPT-NeoX framework, Apache 2.0

ReasoningCodingVisionFunction callingTool useAgentic

-339.0

Elo · rank #435

Parameters

20.6B

Active params

20.6B (dense)

Context

2K tokens

Architecture

Decoder-only Transformer with RoPE and parallel attention+FF blocks (GPT-NeoX library, 44 layers)

License

Apache-2.0

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ARC-ChallengeReasoning38.0%#152

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning72.3%#43

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning31.6%#127

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning24.3%#281

best: GPT-5.6 · 94.6%

GSM8KMath5.5%#184

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning53.5%#147

best: Claude 3 Opus · 95.4%

IFEvalReasoning25.9%#145

best: Gemma 4 26B A4B · 98.5%

LAMBADAReasoning72.0%#13

best: GPT-3 175B · 86.4%

MMLU-ProKnowledge11.6%#180

best: Claude Fable 5 · 91.5%

MMLUKnowledge29.0%#271

best: OpenAI o3 · 92.9%

OpenBookQAReasoning29.0%#50

best: Claude 1 · 90.8%

TriviaQAKnowledge25.9%#49

best: Sarvam-1 (2B) · 90.6%

TruthfulQAKnowledge31.6%#104

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

WinoGrandeReasoning66.1%#114

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

13.1 GB

VRAM @ FP16

43 GB

Fits on (Q4)

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

Throughput data unavailable.

Quantizations

GGUF

GPT-NeoX family

Elo progression across releases

GPT-NeoX-20BFeb 2022-339.0

Resources

HFEleutherAI/gpt-neox-20b GHEleutherAI/gpt-neox DOCSblog.eleuther.ai/announcing-20b

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