Chinese AI startup MiniMax has launched its M3 coding model, featuring a one-million-token context window and significant efficiency gains. The company claims M3 outperforms OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro on a key coding benchmark. However, the model is not fully open source. The launch comes as MiniMax prepares for an IPO on Shanghai's Star Market amid strong revenue growth.
MiniMax M3 coding model launch
- ▪Chinese AI startup MiniMax launched its flagship M3 model, designed for complex coding tasks and automated workflows
- ▪The M3 model can process up to 1 million tokens of data at once, five times more than its predecessor, M2.7
- ▪The M3 model underpins Mavis, MiniMax's recently launched multi-agent platform, as part of its agentic AI strategy
Open-weight versus open-source AI debate
- ▪MiniMax has not released M3’s training code or inference operators, meaning the model cannot be classified as fully open source
- ▪The M3 launch positions MiniMax against competitors in the open-weights AI space like Nvidia’s Nemotron 3 Ultra and Moonshot AI’s Kimi
M3 computational efficiency improvements
- ▪M3's redesigned architecture reduces computational requirements to as little as one-twentieth of previous levels
- ▪The efficiency improvements of the M3 model have slashed inference costs while boosting response speeds
M3 benchmark performance results
- ▪In a company-cited test, the M3 model successfully identified how to optimize software running on Nvidia’s Hopper chips
- ▪MiniMax did not disclose the size of the M3 model or the computing infrastructure used for its training
- ▪MiniMax claims its M3 model outperformed OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro on the SWE-Bench Pro coding benchmark
MiniMax IPO preparations
- ▪MiniMax's 2025 revenue reached US$79 million, a 159% increase year-on-year
- ▪The M3 launch is MiniMax's first major product release since the company began preparations for an IPO on Shanghai’s Star Market
- ▪The planned Shanghai listing will complement MiniMax's current listing in Hong Kong
Story comments
Loading comments…