MiniMax is one of the Chinese companies developing their own generative artificial intelligence models. Based in Shanghai, it combines research into foundation models with the commercialization of services directly accessible to users. Its positioning is built on multimodality: producing or interpreting different types of content, rather than limiting itself to written conversation. Its website, minimax.io, presents its technologies and provides access to its developer offering.
A company born out of AI research
Established in 2021, MiniMax was co-founded by Yan Junjie, formerly a technical leader at SenseTime, a Chinese computer vision specialist. It belongs to the generation of young companies that emerged in China around large models, in an ecosystem where startups operate alongside the laboratories of established technology groups.
MiniMax has chosen to develop both the technical building blocks and the products that use them. This dual approach allows it to offer its models to third parties while also testing them against real-world usage in its own applications. The company also targets audiences outside China, particularly with its video creation tools and its services for conversations with virtual characters.
From foundation models to content creation
MiniMax’s activities span several technology families. Its language models support text generation, dialogue and reasoning. Its work also covers image understanding, speech synthesis and video generation. These capabilities are offered in different forms: consumer interfaces, APIs enabling integration into other software and, for some models, the release of weights.
Hailuo AI is its showcase for generative video. The service notably allows users to create short clips from text prompts or images. Talkie explores another market: interactions with conversational characters whose identities and exchanges users can customize. These products serve different purposes, ranging from a creative tool to interactive entertainment.
The release of MiniMax-01, followed by MiniMax-M1 in 2025, also illustrates its strategy of opening up certain models. Providing access to the weights allows technical teams to evaluate, adapt or deploy them, subject to their licenses and the necessary computing resources. This approach complements hosted API access without making all of MiniMax’s services open.
What next?
For MiniMax, the challenge is to turn this technological diversity into sustained usage and recurring revenue. Video and voice open up opportunities in creative work, communication and digital interfaces, but also entail computing costs and reliability requirements. Its international development will depend as much on model quality as on content rights management, data protection and moderation. In a highly competitive market, integration into work tools will be an important test, beyond the initial appeal of demonstrations.