MiniMax debuts AI model built for long and complex coding tasks

Chinese artificial intelligence start-up MiniMax has unveiled its latest flagship AI model, M3, designed to anchor the company’s push into coding agents and automated workflows. The Shanghai-based company said on Monday that the model’s redesigned architecture reduced computational requirements to as little as one-twentieth of previous levels, slashing inference costs while boosting response speeds. Notably, MiniMax said M3 could process up to 1 million tokens of data at once – five times more than its predecessor, the M2.7. This allowed the model to handle long, complex programming projects. In one benchmark test cited by the company, M3 successfully figured out how to optimise software running on Nvidia’s Hopper chips. According to the company’s WeChat post, MiniMax-M3 beat rival models like OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro on the major coding benchmark SWE-Bench Pro, showing that the model was highly capable at handling software engineering and automated tasks. The company did not disclose the size of the model or the computing infrastructure it used for training. The debut of MiniMax-M3 marked the company’s first major product launch since the firm formally began preparations for an initial public offering on Shanghai’s tech-heavy Star Market, which will complement its current listing in Hong Kong. The dual-market push comes as investors intensify scrutiny over whether AI companies can translate technical advances into sustainable revenue growth, with coding assistants and autonomous agents being widely considered the most commercially promising applications of generative AI. Industry heavyweights – including Anthropic and OpenAI in the US, as well as Alibaba Group Holding and DeepSeek in China – have all accelerated the roll-out of agent-based products in recent months. Meanwhile, the M3 model serves as the bedrock for MiniMax’s broader agentic AI strategy. The firm recently introduced Mavis, a multi-agent system that allows several AI agents to operate simultaneously on a single device, each delegated to distinct tasks. These systems go beyond basic chatbots. Instead of just answering questions, they are designed to act like AI project managers – handling multi-step software tasks and running entire projects with minimal human supervision. Speaking at an earnings briefing in March, CEO Yan Junjie said MiniMax was not aiming for blanket dominance across every AI metric. Instead, the company was focused on building highly differentiated capabilities, he said. To drive adoption, MiniMax launched a seven-day promotional campaign for its M3 application-programming-interface (API) service, offering a 50 per cent discount for usage of up to 512,000 tokens. While the model’s capabilities are currently accessible via API, the company has yet to fully release its underlying inference operators and training code. The release comes amid intensifying competition in the open-source AI market. Nvidia unveiled Nemotron 3 Ultra on Sunday, which it described as the most capable US open-weights model to date, while Chinese developers including Moonshot AI’s Kimi and DeepSeek have continued to push the frontier of open-source model performance. MiniMax has emerged as one of China’s most commercially successful AI companies. Last week, it partnered with Ant Group’s Alipay, becoming one of the first large-model developers to integrate with the platform’s AI payment infrastructure. The partnership gives MiniMax access to global payment, subscription and settlement capabilities – a major step towards monetising AI services beyond consumer chat apps. Ant Group is the fintech affiliate of Alibaba, owner of the South China Morning Post. According to its first earnings report released in March, MiniMax’s revenue reached about US$79 million in 2025, up 159 per cent year on year. Overseas revenue accounted for roughly 73 per cent of total sales. Annual recurring revenue exceeded US$150 million by February this year, while registrations for its open platform products spiked more than fourfold between December and February.