MiniMax AI Model M1 200 Times Cheaper Than GPT-4

Coin WorldWednesday, Jun 18, 2025 1:28 pm ET
1min read

MiniMax, a Shanghai-based AI company, has introduced its latest AI model, M1, which it claims is significantly cheaper to train than OpenAI's GPT-4. According to MiniMax, the training cost for M1 was approximately $534,700, which is nearly 200 times less than the estimated training cost of GPT-4, which industry experts suggest exceeded $100 million. This cost difference could potentially disrupt the AI industry, as it challenges the high costs associated with training advanced AI models.

MiniMax's claims, if verified, could have significant implications for the AI industry. The high cost of training and running AI models has been a major barrier for many companies, and MiniMax's M1 could potentially lower this barrier. This could lead to increased competition in the AI industry, as more companies may be able to afford to develop and deploy advanced AI models. However, it is important to note that MiniMax's claims have not yet been independently verified, and it remains to be seen whether developers will embrace M1 or respond more tepidly.

MiniMax is backed by some of China's largest tech companies, including Tencent and Alibaba. The company has a range of products, including MiniMax Chat, graphic generator Hailuo AI, and avatar app Talkie. MiniMax claims tens of millions of users across 200 countries and regions, as well as 50,000 enterprise clients. However, there is little public information about the company's CEO, Yan Junjie, or the number of people who work at the company.

Geopolitical and national security concerns have also lessened the enthusiasm of some Western businesses to deploy Chinese-developed AI models. Some experts have questioned the accuracy of MiniMax's claims about the amount and type of computer chips it used to create M1. Additionally, all Chinese-produced models have to comply with Chinese government-mandated censorship rules, which means that they can wind up producing answers to some questions that are more aligned to Chinese Communist Party propaganda than generally-accepted facts.

Despite these concerns, MiniMax's M1 has a significant selling point: it is available for free through an API that MiniMax runs. Developers can also download the entire model for free and run it on their own computing resources. If MiniMax's capabilities are what the company claims, it will no doubt gain some traction. The model also has a "context window" of 1 million tokens, which is equivalent to about seven or eight books or about one hour of video content. This means it can take in more data than some of the top-performing models, such as OpenAI's o3 and Anthropic's Claude 4 Opus, which both have context windows of only about 200,000 tokens.