Model
Model explorer

LG EXAONE 3.0 7.8B Instruct

OPEN
LG AI Research · EXAONE 3.0 family · released Aug 7, 2024

LG's first public bilingual (KO/EN) open-weights instruct model, pretrained on 8T tokens.

ReasoningCodingVisionFunction callingTool useAgentic
493.8
Elo · rank #304
Parameters
7.8B
Active params
7.8B (dense)
Context
4K tokens
Architecture
Dense decoder-only Transformer, 32 layers
License
EXAONE AI Model License Agreement 1.1 - NC
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AlpacaEvalHuman preference45.0%#22
best: Tülu 2+DPO 70B · 95.1%
ARC-ChallengeReasoning63.7%#74
best: Llama 3.1 405B · 96.9%
Arena-HardHuman preference46.8%#42
best: Qwen3-235B-A22B (Non-Thinking) · 96.1%
BIG-Bench HardReasoning41.4%#113
best: ERNIE 4.5 300B-A47B · 94.3%
GPQA DiamondReasoning26.6%#271
best: GPT-5.6 · 94.6%
GSM8KMath79.8%#83
best: Llama 3.1 405B · 96.8%
HumanEvalCoding72.0%#76
best: Claude Opus 4.5 · 99.4%
IFEvalReasoning72.1%#109
best: Gemma 4 26B A4B · 98.5%
KoMT-BenchHuman preference8.92#1
best: this model · 8.92
LogicKorHuman preference8.62#11
best: GPT-4o · 9.33
MATH-500Math34.4%#154
best: GPT-5 · 99.4%
MBPPCoding47.4%#74
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
MMLU-ProKnowledge27.4%#167
best: Claude Fable 5 · 91.5%
MT-BenchHuman preference9.01#10
best: Hunyuan-Large (A52B) · 9.4
WildBenchHuman preference48.2%#6
best: Mistral Small 3.2 (24B) · 65.3%
Run it locally
VRAM @ Q4
5 GB
VRAM @ FP16
16 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 Q4 · AWQ
EXAONE 3.0 family
Elo progression across releases
LG EXAONE 3.0 7.8B InstructAug 2024493.8
Resources
HFLGAI-EXAONE/EXAONE-3.0-7.8B-InstructGHLG-AI-EXAONE/EXAONE-3.0
API price weights · each benchmark row carries its own source badge (see methodology)