Model

LG EXAONE 4.0 32B (Reasoning)

OPEN

LG AI Research · EXAONE 4.0 family · released Jul 15, 2025

Opt-in reasoning (enable_thinking=True) mode of Korea's first open-weight hybrid model; highest 32B-class score on math/coding/knowledge benchmarks.

ReasoningCodingVisionFunction callingTool useAgentic

1549.8

Elo · rank #111

Parameters

32B

Active params

32B (dense)

Context

131K tokens

Architecture

Dense hybrid-attention transformer, 64 layers, local:global attention 3:1, GQA (40Q/8KV), QK-Reorder-Norm; single checkpoint with toggleable non-reasoning/reasoning modes

License

EXAONE AI Model License Agreement 1.2 (NC)

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AIMEMath85.3%#67

best: GPT-5.2 · 100.0%

BFCL v3Agents63.9%#23

best: Hunyuan-A13B · 78.3%

GPQA DiamondReasoning75.4%#102

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning10.5%#88

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning36.3%#59

best: MiniMax M3 · 83.0%

IFEvalReasoning83.7%#74

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding72.6%#63

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMLU-ProKnowledge81.8%#47

best: Claude Fable 5 · 91.5%

MMLU-ReduxKnowledge92.3%#18

best: Qwen3.7-Max · 95.0%

MMMLUKnowledge85.6%#17

best: Gemini 3.1 Pro · 92.6%

tau-benchAgents62.8%#12

best: Claude Opus 4.1 · 82.4%

τ²-Bench TelecomAgents17.3%#38

best: Claude Opus 4.6 · 99.3%

Run it locally

VRAM @ Q4

20 GB

VRAM @ FP16

64 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

FP8 · GGUF Q4

EXAONE 4.0 family

Elo progression across releases

LG EXAONE 4.0 32B (Non-reasoning)Jul 20251238.1 LG EXAONE 4.0 32B (Reasoning)Jul 20251549.8

Resources

HFLGAI-EXAONE/EXAONE-4.0-32B GHLG-AI-EXAONE/EXAONE-4.0

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