Naver Cloud Releases 14B Parameter AI Model for Commercial Use

Seoul, South Korea - Naver Cloud Corp. announced the release of a 14-billion-parameter artificial intelligence model as open-source software for commercial use.

The company stated that the model, known as HyperCLOVA X SEED 14B Think, was developed utilizing proprietary technology.

Naver Cloud stated that it trained the model at one percent of the cost of comparable global open-source models of a similar size.

The company said it achieved the cost reduction through parameter pruning and knowledge distillation techniques.

Naver Cloud stated that the model performed comparably to other 14-billion parameter models and some 32-billion parameter models in evaluations of Korean language capabilities, coding, and mathematical reasoning.

The model is a smaller version of HyperCLOVA X THINK, which the company announced on June 30, according to Naver Cloud.

The company said it designed the model for integration into commercial services and AI agent applications.

Naver Cloud said three AI models it released as open-source in April had been downloaded more than one million times by July.

The company said more than 50 derivative models have been developed from those releases.

The company stated that it evaluated the model using benchmark tests, including KMMLU, KoBALT, HAERAE, MATH500, HumanEval, MBPP, and LiveCodeBench.

"We were able to build a more effective reasoning model in terms of cost and performance at the same scale, based on high-performance flagship models continuously upgraded with proprietary technology," said Sung Nak-ho, Naver Cloud’s Chief Technology Officer for Hyperscale AI.