general

NAVER's HyperCLOVA X SEED: 32B Beats GPT-4o

NAVER unveils HyperCLOVA X SEED, a 32-billion parameter language model that surpasses GPT-4o in benchmark performance, marking a significant advancement in

Someone found that NAVER’s new HyperCLOVA X SEED models punch way above their weight class - the 32B “Think” variant apparently outperforms GPT-4o on several benchmarks while being significantly smaller.

Two versions just dropped:

The interesting bit: these are open weights, so people can actually run them locally or fine-tune them. The 8B model is small enough to run on consumer hardware while still handling text, images, and audio.

Full collection at https://huggingface.co/collections/naver-hyperclovax/hyperclova-x-seed

Pretty notable since most competitive models this capable are either closed-source or much larger. Worth checking out if you’re tired of API costs or need something that works offline.