MiniMax is building the world's largest open LLM at 2.7 trillion parameters

Chinese AI startup MiniMax is developing a large language model with 2.7 trillion parameters, Reuters reports, citing a source familiar with the matter. It is the largest open-weight AI model released by a Chinese company, and possibly the largest open model in the world.

How big is MiniMax's model, really

At 2.7 trillion parameters, MiniMax's upcoming model ranks among the largest language models known today by stated size. For comparison, DeepSeek's DeepSeek V3 (671 billion parameters) hit the open AI model market like a bombshell when it launched. Yet it has almost four times fewer parameters than the model MiniMax is building.

• Developer: Chinese AI startup MiniMax
• Stated model size: 2.7 trillion parameters
• Source: Reuters, citing a source familiar with the matter
• The model is billed as the largest open AI model released by a Chinese company

Who is MiniMax, and why it matters

MiniMax is one of China's so-called "six tigers" of AI startups. Based in Shanghai, it builds its own large language, voice, and video models under the abab and MiniMax-01/M-series brands. The company has already released open models and API services for text, speech, and video generation, and it competes at home with DeepSeek, Alibaba (the Qwen model family), Moonshot AI (Kimi), and Zhipu AI. Building a model at the 2.7 trillion-parameter scale takes substantial compute, and it signals MiniMax's serious ambition to secure a place among the world's top AI labs, not just in the Chinese domestic market.

Why the race for model size has heated up again

Language models at this scale are almost always built on a MoE (mixture-of-experts) architecture, in which only a fraction of the total parameters handles any given request. That lets developers grow a model's total size without a proportional increase in inference compute costs. The parameter race among Chinese AI labs has become one of the main fronts in their competition with American labs in the open-weight market. After DeepSeek's open models impressed the industry with their balance of quality and training cost, interest in Chinese open models surged among developers worldwide looking for alternatives to the closed APIs from OpenAI, Anthropic, and Google.

Open weights mean that once the model is published, anyone can download it, fine-tune it for their own tasks, and run it on their own infrastructure without depending on a specific vendor's API. That sets Chinese labs fundamentally apart from most leading American developers, which keep the weights of their flagship models closed.

What this means

If Reuters' report is confirmed, MiniMax will cement its status as the developer of the world's largest open language model. It is another sign that Chinese AI labs, not American ones, are currently setting the pace in the race to scale open-weight models.