China's cheaper AI tokens a double-edged sword for Asian businesses

Chinese AI models are gaining traction in Asia due to their lower token costs, which could make large-scale AI adoption more affordable for businesses in price-sensitive markets like India and Southeast Asia. Models from companies such as MiniMax and Moonshot charge between US$2 and US$3 per million output tokens, far below Western competitors like Google’s Gemini 3.5 Flash (US$9), Anthropic’s Claude Sonnet 4.5 (US$15), and OpenAI’s GPT 5.5 (US$30), according to pricing documents and a Financial Times report. Token costs accumulate based on input and output usage—input tokens come from prompts, while output tokens generate responses, typically costing more. A small sales team of 50 employees could consume around 450 million tokens monthly, leading to annual costs of US$38,000 with GPT 5.5, roughly two to three times higher than Chinese alternatives. The lower pricing stems from efficient model designs, reduced infrastructure costs, government subsidies, and aggressive pricing strategies, experts told CNA. This affordability is driving adoption among startups and enterprises, particularly in sectors like call centers, software development, and e-commerce. Companies such as Airbnb, Thinking Machines Lab (founded by former OpenAI CTO Mira Murati), and AI Singapore have already integrated Alibaba’s Qwen models into their operations. However, cheaper tokens come with trade-offs, including potential compromises in quality, latency, trust, data security, and geopolitical risks. Observers note that while cost savings are significant, businesses must weigh these factors against performance needs. Token costs multiply as AI systems perform complex tasks like planning, searching, and verifying information, making pricing a critical factor in scalability. The pricing model disproportionately affects businesses and developers integrating AI into products, apps, and workflows, as they pay per usage rather than fixed fees. This model is increasingly common because enterprises use AI at scale—handling millions of customer chats, coding requests, and research tasks—where token consumption can rise sharply. Experts emphasize that while Chinese AI models offer a cost-effective entry point, long-term viability depends on balancing affordability with operational reliability.