Chinese AI upstart MiniMax released a new large language model, joining a slew of domestic peers inspired to surpass DeepSeek in the field of reasoning AI. The Shanghai-based company touted the efficiency of its new MiniMax-M1 model in handling complicated productivity tasks, claiming it outdoes all closed-source competitors from China in a statement. In several benchmarks presented by MiniMax, M1 also scored higher than DeepSeek’s latest R1-0528 model. M1 supports a context length of a million tokens, eight times the size of DeepSeek R1. Context length helps AI systems process more information simultaneously. Under some circumstances, MiniMax’s model requires only about 30% of the resources that DeepSeek would, the company said. Bloomberg hasn’t independently verified MiniMax’s claims. The company used large-scale reinforcement learning to train M1, employing 512 Nvidia Corp. H800 GPUs with a rental cost of $534,700, according to the statement. Backed by China’s biggest internet companies, Tencent Holdings Ltd. and Alibaba Group Holding Ltd., MiniMax belongs to an elite group of a half dozen domestic AI startups known as the Little Dragons. Together, they have raised billions of dollars in venture funding over the past year — though the rise of DeepSeek forced the majority of the group to cut back or terminate fundamental research and focus more on applications.

Full report : Shanghai-based MiniMax open sources MiniMax-M1, a model for complicated productivity tasks that supports 1M input tokens and it says beats DeepSeek’s R1-0528.