Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size

Tencent has unveiled Hy3, an open-source language model featuring 295 billion parameters. This model utilizes a mixture-of-experts architecture, activating only 21 billion parameters at any one time. Hy3 claims to match models two to five times its active size while halving the hallucination rate to 5.4%. This performance highlights significant advancements in language model design. The launch of Hy3 is a pivotal moment in AI, offering an efficient open-source model that could reshape language application development in both the Arab market and globally.