Model

NAVER HyperCLOVA X

CLOSED

NAVER · HyperCLOVA X family · released Aug 24, 2023

NAVER's sovereign Korean-first LLM (~204B), successor to 2021 HyperCLOVA; served via CLOVA Studio/CLOVA X, API/service only.

ReasoningCodingVisionFunction callingTool useAgentic

491.0

Elo · rank #305

Parameters

204B

Active params

204B (dense)

Context

32K tokens

Architecture

Dense transformer decoder (closed, details undisclosed)

License

Proprietary

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning53.3%#13

best: OLMo 3-Think 32B · 88.2%

BIG-Bench HardReasoning53.5%#91

best: ERNIE 4.5 300B-A47B · 94.3%

GSM8KMath81.7%#74

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning86.0%#31

best: Claude 3 Opus · 95.4%

HumanEvalCoding55.5%#111

best: Claude Opus 4.5 · 99.4%

LogicKorHuman preference7.76#18

best: GPT-4o · 9.33

MATH-500Math20.2%#182

best: GPT-5 · 99.4%

MBPPCoding33.4%#93

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

MMLUKnowledge68.0%#161

best: OpenAI o3 · 92.9%

PIQAReasoning86.4%#7

best: GPT-4o mini · 93.1%

TriviaQAKnowledge63.6%#31

best: Sarvam-1 (2B) · 90.6%

TruthfulQAKnowledge69.5%#5

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

WinoGrandeReasoning79.6%#39

best: PaLM 2 · 90.9%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

HyperCLOVA X family

Elo progression across releases

NAVER HyperCLOVA XAug 2023491.0

Resources

HFnaver-hyperclovax DOCSclova.ai/en/clova-studio APIAvailable via NAVER API

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