Model

LG EXAONE 3.5 32B Instruct

OPEN

LG AI Research · EXAONE 3.5 family · released Dec 9, 2024

Flagship of the 2.4B/7.8B/32B EXAONE 3.5 family; 32K context, strong instruction-following.

ReasoningCodingVisionFunction callingTool useAgentic

911.3

Elo · rank #232

Parameters

32B

Active params

32B (dense)

Context

32K tokens

Architecture

Dense Transformer, 64 layers, GQA (40 Q-heads / 8 KV-heads)

License

EXAONE AI Model License Agreement 1.1 - NC

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AlpacaEvalHuman preference60.6%#13

best: Tülu 2+DPO 70B · 95.1%

ARC-ChallengeReasoning91.7%#20

best: Llama 3.1 405B · 96.9%

Arena-HardHuman preference78.6%#28

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

BIG-Bench HardReasoning75.3%#41

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning39.7%#221

best: GPT-5.6 · 94.6%

GSM8KMath91.9%#32

best: Llama 3.1 405B · 96.8%

HumanEvalCoding87.2%#36

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning81.7%#87

best: Gemma 4 26B A4B · 98.5%

KoMT-BenchHuman preference8.05#2

best: LG EXAONE 3.0 7.8B Instruct · 8.92

LogicKorHuman preference9.06#5

best: GPT-4o · 9.33

MATH-500Math70.5%#89

best: GPT-5 · 99.4%

MBPPCoding81.8%#11

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

MMLU-ProKnowledge46.4%#141

best: Claude Fable 5 · 91.5%

MMLUKnowledge78.3%#104

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.51#25

best: Hunyuan-Large (A52B) · 9.4

Run it locally

VRAM @ Q4

20 GB

VRAM @ FP16

62 GB

Fits on (Q4)

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

Throughput data unavailable.

Quantizations

GGUF Q4 · AWQ

EXAONE 3.5 family

Elo progression across releases

LG EXAONE 3.5 32B InstructDec 2024911.3

Resources

HFLGAI-EXAONE/EXAONE-3.5-32B-Instruct GHLG-AI-EXAONE/EXAONE-3.5

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