MiniMax Prepares 2.7 Trillion Parameter Open-Source Model — One of the Largest AI Models in History

Introduction: A New Chapter in Open-Source AI

Chinese AI startup MiniMax is developing a new large language model (LLM) with a staggering 2.7 trillion parameters — and plans to release it as open source later this year. This would be one of the largest open-source AI models ever released, both in China and globally.

According to The Information, which cites two sources close to the project, the new MiniMax model is internally codenamed M3 Pro, though the name may change before the official launch. The initial release is expected around Q3 2026.

M3 Pro: Technical Specifications and Scale

MiniMax's current flagship model, M3, operates with 428 billion parameters — an impressive figure in its own right. However, the M3 Pro at 2.7 trillion parameters surpasses its predecessor by more than 6x and exceeds any existing Chinese AI model.

For comparison: Meta's largest Llama 3 has 405 billion parameters, DeepSeek-V3 has 671 billion, and GPT-4's exact parameter count is undisclosed but estimated around 1.7 trillion. The M3 Pro at 2.7 trillion leaves them all behind.

Competition Between Chinese and Western AI Companies

This move by MiniMax is part of an ambitious strategy by Chinese AI companies competing not only in the domestic market but also globally — directly challenging companies like OpenAI, Google, Anthropic, and Meta. Chinese AI startups including Zhipu AI, DeepSeek, Moonshot AI, and 01.AI have all released open-source models, making the Chinese AI ecosystem one of the most dynamic in the world.

Geopolitics and Export Controls: A Complex Context

MiniMax operates in a complex geopolitical environment. The Chinese government is reportedly considering AI export controls, which would mirror similar restrictions imposed by the US. These controls could affect which hardware MiniMax can access for training and inference.

Impact on the Open-Source Ecosystem

If MiniMax succeeds in releasing M3 Pro as open source, it could be a game-changer for AI. The open-source community would gain access to a model that rivals the most powerful proprietary systems. This could accelerate AI research globally, particularly in regions with limited access to cutting-edge proprietary models.

Challenges and Risks

The biggest challenge is training — a 2.7 trillion parameter model requires enormous computational resources. Even if training is successful, inference costs remain significant. And distribution as open source raises questions about safety and misuse.

Conclusion

If MiniMax succeeds in launching M3 Pro as open source, it could become one of the defining milestones of 2026 in AI — a model that challenges the notion that the largest, most capable models must remain behind closed doors.