A Chinese large language model (LLM) with frontier-level performance is making significant inroads into the domestic AI agent market, offering prices as low as one-twentieth of comparable U.S. models. Following DeepSeek, Minimax has begun targeting the South Korean market, prompting local companies to consider switching to Chinese AI solutions due to cost concerns.
According to the IT industry on the 28th, Chinese AI startup Minimax has launched its services in South Korea. The company has set the output token price for its latest model, M2.7, at $1.20 per million tokens. This is just 8% of Anthropic's Claude Sonnet 4.6 price of $15 per million tokens and 12% of GPT-4o's $10 per million tokens. Compared to Claude Opus 4.6, which costs $25 per million tokens, Minimax's price is approximately one-twentieth.
Minimax, based in Shanghai, went public on the Hong Kong Stock Exchange in January. Major investors include Alibaba, Tencent, and Hillhouse Investment, with the company's valuation reaching $4 billion last year. The M2.7 model is optimized for agent coding and tool usage, and API access is available through platforms like OpenRouter.
DeepSeek is also following this trend. The company has permanently reduced the API price for its flagship model, V4-Pro, by 75%, lowering the output token price to $0.87 per million tokens. This reduction brings the input token price for DeepSeek V4-Pro down to about $0.40 per million tokens, a quarter of its previous cost. The DeepSeek V4 Flash model is now priced at approximately $0.14 per million tokens, establishing itself as one of the lowest-cost high-performance models in the global market.
The strategic focus on aggressive pricing by Chinese AI companies is driven by the structural characteristics of the AI agent market. A paper published by Stanford's Digital Economy Lab, titled "Token Consumption Analysis and Prediction for Agent Coding Tasks," indicates that agent tasks can consume up to 1,000 times more tokens than code inference or chat-based tasks. Agents must re-input the entire context of previous conversations at each step, leading to a phenomenon known as "context snowballing" as tasks lengthen, according to the research team.
While API costs for simple chatbot usage may be in the range of a few cents, transitioning to AI agents for task automation can result in costs skyrocketing by hundreds of times.
As cost pressures mount, domestic startups are also responding. Some companies developing and operating AI applications are in the process of replacing GPT or Claude-based systems with DeepSeek.
An AI developer stated, "In a situation where the revenue structure is uncertain, high token prices lead to financial burdens, making the adoption of Chinese AI essential. While it may not be a complete replacement, most companies are incorporating DeepSeek and similar solutions to some extent."
It is reported that large IT service companies are also testing Chinese LLMs like DeepSeek and Minimax internally to evaluate their performance and cost-effectiveness.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.