MiniMax, a tech company, recently announced that since its establishment at the end of 2021, it has built trillion-parameter MoE text, voice, and image large models, and launched native applications such as Xingye and Hailuo AI, while also providing an open platform API service for enterprises and developers.
In the past 996 days, MiniMax's large models have interacted with end-users an average of 3 billion times per day, processed over 3 trillion text tokens, and generated 20 million images and 70,000 hours of voice. These interactions showcase the company's rapid progress and widespread influence in the field of artificial intelligence.
IO, the founder of MiniMax, emphasized that despite significant progress, the company still has a long way to go to achieve the goal of 100% intelligent connectivity for the global population. To realize this goal, MiniMax will continue to focus on technological innovation, especially in reducing model error rates, supporting unlimited length input and output, and multi-modal interactions.
During the Partner Day, MiniMax released the video model abab-video-1, the music model abab-music-1, and updated the voice model abab-speech-1. Additionally, MiniMax plans to release a multi-modal model, abab7, using MoE + Linear Attention technology in the coming weeks.
Furthermore, MiniMax has introduced advanced voice and video models, including voice models that support multiple languages and emotional expressions, as well as video models with high compression rates and diverse styles. These models will be available to users and developers through the company's open platform and products like Xingye and Hailuo AI.