MiniMax AI: A Tough Competitor
Chinese technology company, which has emerged as a leading force in the field of artificial intelligence powered video, music and speech generation. With its innovative AI models and advanced algorithms, MiniMax is revolutionizing the way content is created and consumed.
In terms of daily processing volume, MiniMax algorithm in AI is the biggest generative multi-mode universal large model enterprise in China. MiniMax AI launched native applications such as the productivity tool Conch AI and the immersive AI content community Xingye.
MiniMax’s open platform provides enterprises and developers with safe, flexible and reliable API services to help quickly build AI applications.
Three billion user interactions per day take place between the MiniMax model and consumers worldwide, involving the following:
- Processes more than 3 trillion text tokens daily, which is the equivalent of 3,000 life experiences.
- Generates 20 million images daily, which is the equivalent of 400 Forbidden City Palaces’ worth of paintings.
- Performs 70,000 hours of speech synthesis daily, which is the equivalent of reading 7,000 books in a single day.
Currently, MiniMax’s big model engages in 3 billion daily interactions with end users. What is meant by 3 billion times? This involves producing 20 million photos daily, 70,000 hours of voice every day, and processing over 3 trillion text tokens daily. MiniMax’s daily contact time on ChatGPT was approximately 3% a year ago, but it has now increased to over 50%.
MiniMax AI: China’s Frontier in AI Video and Audio Generation
What is meant by three trillion text tokens? It is the same as going through 3,000 life experiences in a single day.
Users who have grown up from all across the world are behind these 3 billion connections. They are diverse, creative, and vibrant, and they come from all across nations, regardless of gender, age, or youth. MiniMax AI aims using cutting-edge technology to surprise people, and serve their creative content making goals.
As of right now, the Conch AI web version and the open platform are both home to the MiniMax music and video models. In order to enable creativity and imagination through technology, they intend to collaborate with additional ecological partners in the future to investigate the landing possibilities of music and video models.
Founder of MiniMax AI is Yan Junjie. According to his “Intelligence with Everyone” concept, this is the only and most effective method of achieving AGI (Artificial general intelligence). MiniMax, as of now has: speech model Abab-speech-1 and launched the video model Abab-video-1 and music model Abab-music-1.
Abab-video-1 is similar to the texture of movies and offers features including a high compression rate, good text response, support for native high-resolution and high-frame rate videos, and a variety of styles by Minimax’s Hailuoai.
Users can experience creating their own unique creative videos by logging into the Hailuo- Conch AI web version.
Abab-music-1: The process of creating and recording music is made much simpler, which offers versatile end-to-end music production. It may be used to synthesis a wide range of musical genres, including vocal and accompaniment works as well as pure music.
Abab-speech-1: Many languages, including Cantonese, Korean, Spanish, Japanese, etc., are supported too . The generated phrases exhibit a high degree of humanization, with subtle and organic emotional shifts.
As of now, MiniMax AI is not a fully subscription-based service. It offers its video and audio generation features for free, requiring users to register through their website. For video generation, MiniMax currently supports text-to-video capabilities, which is available for free trial. This model, launched recently, allows users to create short videos from text prompts, though it still carries some limitations, such as a maximum video length of 5 seconds with watermarks.
Above is a test video Prompt used: ” A vibrant anime scene of a boy & girl wearing headphones under a starry sky, as they walk gently hand in hand. A river flowing in the backdrop and a tree with leaves swaying in wind”
Commercialization strategy of MiniMax AI is still evolving, with some focus on serving business clients. The company currently has over 2,000 clients using its open platform and plans to introduce further iterations of its video generation model soon. However, for now, it remains in a development phase, emphasizing technical improvements over commercialization
MiniMaxi AI, furthermore, in the upcoming weeks will release the multimodal model Abab 7 that makes use of MoE + Linear Attention technology. Implementation of this newest model technology based on MoE (hybrid expert architecture) and Linear Attention will commit to push the limits of human intelligence and accelerating AI development.
It will facilitate the training of enormous amounts of data, considerably increase practicality and response time, and drastically lowers the expenses associated with large models. This new design will save more than 90% of the cost as compared to the standard Transformer architecture, and the benefit becomes more apparent as the data sequence length increases.
The new generation Abab model doubles its efficiency when compared to the capabilities of the same generation model of GPT-4o, and the benefit becomes more noticeable as the length rises.
The motto of MiniMax is “Intelligence in All of Us”, claiming that the users are the ones who assists the algorithm in self learning, rather than their technology, and can improve intelligence by more involvement and input.
To genuinely achieve “Intelligence with Everyone”, their aim is to collaborate with maximum users. Yan Junjie says, “We cannot have greater intelligence until a diverse range of people participate and provide input to our algorithm.”
ALSO READ BELOW ARTICLES ON AI:
ALAYA AI: Earn Crypto by Labeling Data? The Future of Work is Here!
Flux AI: Utilizing Smarter & Faster automation for Hardware Design
Skild AI: $300M Funding Round for Revolutionary feats in future Robotics