MiniMax debuts AI model built for long and complex coding tasks
MiniMax launched M3, a flagship model for coding agents, claiming faster processing, lower inference costs and a larger context window.
Intelligence analysis by GPT-5.4 Mini
Shanghai-based MiniMax says M3 is its new flagship model for coding agents and automated workflows. The company claims a redesigned architecture cuts compute needs, speeds responses and expands the model’s ability to handle long programming projects.
MiniMax says it built a new AI model called M3 that is better at handling very long computer coding jobs. It is meant to act like a helper that can keep track of a huge project without forgetting what came earlier.
The company says M3 is faster and cheaper to use than its older version. It also says M3 can look at much more information at once, like a student who can read a much thicker book before answering questions.
MiniMax is using this launch to show it can compete in AI coding tools. It also comes as the company gets ready for a possible stock market listing in Shanghai.
Analysis
What MiniMax launched
MiniMax, a Shanghai-based AI start-up, introduced M3 as its latest flagship model. The company says the model is meant to support coding agents and automated workflows, with a focus on long and complex programming tasks.
What the company claims
MiniMax says M3’s redesigned architecture reduces computational requirements to as little as one-twentieth of previous levels. According to the company, that translates into lower inference costs and faster responses. MiniMax also says M3 can process up to 1 million tokens at once, which it describes as five times the capacity of its predecessor, M2.7. That larger window is meant to help the model work through longer software projects without losing track of earlier context.
The company pointed to a benchmark test in which M3 reportedly figured out how to optimize software running on Nvidia’s Hopper chips. MiniMax also said, in a WeChat post, that M3 beat OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro on SWE-Bench Pro, a major coding benchmark used to gauge software engineering ability.
Why this launch matters
The debut is MiniMax’s first major product release since it formally began preparing for an initial public offering on Shanghai’s tech-heavy Star Market, alongside its existing Hong Kong listing. The announcement suggests the company is trying to show technical progress in a competitive AI market while emphasizing practical use cases tied to coding and automation.
MiniMax did not disclose the model’s size or the computing infrastructure used for training, so the claims remain those of the company rather than independently verified facts.
Key points
- MiniMax unveiled M3 as its newest flagship AI model for coding agents and automated workflows.
- The company says M3 can process up to 1 million tokens at once, five times more than M2.7.
- MiniMax claims the redesigned architecture cuts computational requirements to as little as one-twentieth of previous levels.
- The company said M3 beat OpenAI's GPT-5.5 and Google’s Gemini 3.1 Pro on SWE-Bench Pro, according to its WeChat post.
- The launch is MiniMax’s first major product release since it began preparing for an IPO on Shanghai’s Star Market.
