Reading through a large codebase can take hours. So can sorting through a long contract, a pile of meeting notes or a detailed research report.
MiniMax says its new M3 model is designed to make that kind of work easier.
According to the company, M3 can handle up to 1 million tokens of context. In practical terms, that means it can process much more information at once than many standard AI chatbots. A developer could use it to review a larger section of a software project, while an office worker might ask it to summarize a lengthy report without splitting the file into dozens of smaller parts.
M3 can also work with images and video, not just text. With the right permissions, it can use tools and carry out tasks on a computer, such as searching for information, entering data and organizing material across different apps.
To manage such large amounts of information, MiniMax uses a system known as sparse attention. The idea is simple: instead of giving equal attention to everything, the model tries to focus more on the parts that matter. That can reduce the computing power needed for very long documents.
Still, most of the performance claims so far come from MiniMax’s own tests. Results in everyday use may vary depending on the task, the hardware and how the model is used.
M3 is available through MiniMax’s API and coding tools, and the company has also released its model weights. MiniMax says the model can edit code, use external tools and work through longer, multi-step assignments.
For most people, the technical specifications will matter less than the outcome. The real question is whether M3 can save time, avoid mistakes and help users get through work that would otherwise take an entire afternoon.
