Naver Completes Korean AI Reasoning Model HyperCLOVA X

Seoul, South Korea - Naver Corp. said on Monday that it has completed the development of an artificial intelligence reasoning model called HyperCLOVA X THINK and published a technical report detailing the system's design and performance data.

The company stated that the reasoning model processes user queries by conducting an in-depth internal analysis before generating responses.

This process involves breaking complex problems into smaller components, selecting appropriate tools or functions, and correcting errors, according to the technical report released June 30.

Naver reported that HyperCLOVA X THINK scored higher than comparable domestic reasoning models and major global open-source models in KoBALT-700 benchmark testing.

Seoul National University's Department of Linguistics designed the benchmark to evaluate AI systems on expert-level Korean language comprehension, including syntax, semantics, and pragmatics.

The model also scored higher than major domestic and international open-source models in HAERAE-Bench testing, another Korean language performance evaluation, the company said.

The system demonstrated visual processing capabilities when presented with STEM problems in image format.

In testing with Korean university entrance examination biology questions, the model analyzed ecological succession diagrams and plant community productivity graphs, combining visual recognition with knowledge to select the correct answers from multiple-choice options.

Naver said it plans to release HyperCLOVA X THINK as open-source software. The company's previous open-source model, HyperCLOVA X SEED, exceeded 500,000 downloads within one month of its April release.

The company stated that it utilized proprietary technology in developing the model, including a Peri-LN technique that has been accepted for presentation at the International Conference on Machine Learning 2025.

Naver also stated that it has developed reinforcement learning methods to enhance the model's training efficiency.

The company stated that reasoning models serve as the core technology for AI agent services, which require natural language interaction capabilities between users and AI systems.

Naver said HyperCLOVA X THINK can more accurately understand and perform user instructions in various daily situations.