MiniMax, a Chinese AI startup, announced it will open-source a large language model with 2.7 trillion parameters later this year. The model, internally named M3 Pro, is expected to be released as early as Q3 2026, according to The Information, which cited two sources familiar with the plans. This model would be the largest among Chinese AI models currently available, significantly outpacing MiniMax's current top model, M3, which has 428 billion parameters.
The decision to open-source the model aligns with growing interest in Chinese open-source models, particularly among developers seeking cost-effective solutions for high-volume, less critical tasks. MiniMax competes with other Chinese firms such as Zhipu, DeepSeek, and Moonshot AI. However, recent reports indicate that the Chinese government may seek to impose tighter controls on future model releases.
Source: thedecoder
