South Korean AI startup, Upstage, proudly announced that its generative AI model has achieved the top position on the Open LLM leaderboard hosted on Hugging Face, the world’s largest machine learning platform.

The Open LLM leaderboard on Hugging Face serves as the gold standard for open-source generative AI models. Models are ranked based on their performance across four key metrics: reasoning and common sense, language understanding, hallucination prevention, and fluency.

In a significant achievement over the past month, Upstage’s 30B parameter model secured the first spot, surpassing Meta’s 70B parameter model, LaMDA 2, which was introduced on the same day.

Upstage has now unveiled a 70B parameter model, fine-tuned with more data, successfully reclaiming the top spot globally. This new model achieved an impressive score of 72.3 points on the leaderboard, surpassing the US-based Stability AI’s Stable Beluga 2 model (71.4 points), which had taken the lead following the launch of LaMDA 2.

Notably, Upstage’s latest model has even outperformed the benchmark generative AI model, GPT-3.5, on Hugging Face’s leaderboard. This marks the first time any model has achieved a higher score than GPT-3.5 on the platform.

Moving forward, Upstage aims to commercialize its AI models and venture into the private AI market. Private AI offers specialized solutions that prevent information leakage and hallucination by exclusively learning from internal data of companies. It presents an excellent option for businesses concerned about security issues and hesitant to adopt generative AI.

Kim Sung-hoon, CEO of Upstage, expressed his delight, saying, “We are thrilled to demonstrate our world-leading technology with our generative AI model, surpassing ChatGPT. We will continue to strengthen our position in the domestic and global private AI market, leveraging our outstanding technology.”