ByteDance Trains 10 Trillion Parameter AI Model to Rival Anthropic

ByteDance, the prominent Chinese technology firm, is developing an extensive AI model that could reach up to 10 trillion parameters, potentially tripling the size of Kimi K3, which is currently the largest model released in China. This initiative reflects China's goal to not only catch up but also outpace American laboratories in the realm of advanced AI. The model is currently in the early pre-training phase, which generally lasts between three to six months before it undergoes fine-tuning and possible release. The final size will be confirmed later. While Anthropic does not reveal specific model sizes, estimates suggest its leading Mythos 5 has around 8 trillion parameters, with Fable 5 at approximately 5 trillion. Parameter counts indicate capacity limits, but actual performance relies on data quality and training techniques. Recent models from Moonshot and Alibaba have demonstrated impressive benchmark results, trailing only Anthropic's Fable 5 in certain metrics. Following a temporary ban in June due to security issues, Mythos 5 is now accessible only to authorized organizations. Industry experts indicate that several Chinese laboratories are working on models comparable to Fable 5, with ByteDance notably aiming for the largest scale.