J Studios
Chinese AI startup MiniMax launched a new reasoning large language model called MiniMax-M1 which it claims is even better than DeepSeek’s (DEEPSEEK) upgraded its AI model R1.
MiniMax, which is backed by Tencent (OTCPK:TCTZF) (OTCPK:TCEHY) and Alibaba (NYSE:BABA), claimed the new model outdoes all closed-source competitors from China in complex scenarios for productivity, Bloomberg News reported citing, a company statement which was in Chinese.
M1 also scored higher than DeepSeek’s R1-0528 model.
The company noted that M1 supports a context input of 1 million tokens, eight times that of DeepSeek R1, and the industry’s longest inference output of 80,000 tokens.
Context length helps AI systems process more information simultaneously.
The company noted, according to the Google translated press release, that, for example, when using 80,000 tokens for deep reasoning, only about 30% of the computing power of DeepSeek R1 is needed.
The Chinese company used large-scale reinforcement learning to train M1, using 512 Nvidia (NASDAQ:NVDA) H800 GPUs with a rental cost of $534,700, as per the statement.
MiniMax noted it will share more updates in the coming days. The company’s other products include a video generation tool and an AI companion app, the report noted.
Last month, China’s DeepSeek had upgraded its AI model R1. Earlier this year, DeepSeek rattled the U.S. tech sector after it unveiled its AI model R1 and detailed in a paper how LLMs could be built on tight budgets and improved without human supervision.