Model
Model explorer

GPT-J-6B

OPEN
EleutherAI · GPT-J family · released Jun 9, 2021

6B Pile-trained model built with mesh-transformer-jax, Apache 2.0; the standard open GPT-3 alternative for roughly two years

ReasoningCodingVisionFunction callingTool useAgentic
-391.0
Elo · rank #441
Parameters
6.05B
Active params
6.05B (dense)
Context
2K tokens
Architecture
Decoder-only Transformer with Rotary Position Embeddings and parallel attention+FF blocks (mesh-transformer-jax)
License
Apache-2.0
Languages
1+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning34.0%#158
best: Llama 3.1 405B · 96.9%
ARC-EasyReasoning67.0%#57
best: Phi-3-medium (14B) · 97.7%
BIG-Bench HardReasoning31.9%#126
best: ERNIE 4.5 300B-A47B · 94.3%
GPQA DiamondReasoning24.6%#280
best: GPT-5.6 · 94.6%
GSM8KMath3.0%#188
best: Llama 3.1 405B · 96.8%
HellaSwagReasoning66.1%#134
best: Claude 3 Opus · 95.4%
IFEvalReasoning25.2%#146
best: Gemma 4 26B A4B · 98.5%
LAMBADAReasoning69.7%#19
best: GPT-3 175B · 86.4%
MMLU-ProKnowledge12.4%#178
best: Claude Fable 5 · 91.5%
MMLUKnowledge25.4%#286
best: OpenAI o3 · 92.9%
OpenBookQAReasoning28.8%#51
best: Claude 1 · 90.8%
PIQAReasoning76.5%#68
best: GPT-4o mini · 93.1%
TriviaQAKnowledge17.0%#50
best: Sarvam-1 (2B) · 90.6%
TruthfulQAKnowledge36.0%#93
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
WinoGrandeReasoning65.3%#122
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
4.2 GB
VRAM @ FP16
13.4 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 · Int8
GPT-J family
Elo progression across releases
GPT-J-6BJun 2021-391.0
Resources
HFEleutherAI/gpt-j-6bGHkingoflolz/mesh-transformer-jaxDOCShuggingface.co/docs/transformers/model_doc/gptj
API price weights · each benchmark row carries its own source badge (see methodology)