MiniMax Plans 2.7 Trillion Parameter Model M3 Pro

According to Guandian News, MiniMax plans to launch an internal codenamed large model M3 Pro with 2.7 trillion parameters. This far exceeds publicly known models like GPT-4 (~1.8T) and Gemini Ultra, signaling MiniMax's ambition in ultra-large-scale models. Release date, training cost, and application scenarios have not been announced yet. For those tracking domestic large model development, this means the parameter size competition may escalate. Researchers and engineers focusing on ultra-large model architecture and training optimization should watch whether it solves computational and memory bottlenecks. However, as it is only a preliminary plan without technical details, its impact on actual product deployment remains to be seen.