Chinese AI Models Now Lead Global Token Rankings

The pattern is clear. Chinese AI companies have embraced open-source distribution, aggressive pricing, and rapid iteration. The result is a market where developers worldwide are quietly building on top of Chinese models, drawn by the combination of performance and cost.

"The continued leadership of China's large model call volume reflects the high activity of the domestic AI application ecosystem^✦ and further optimisation of computing costs." - Industry analysis, AI NewsTime, March 2026

Enterprise Implications Across Asia

For CIOs across the Asia-Pacific, the pricing gap creates a genuine strategic question. IDC predicts that by 2027, AI infrastructure costs will run up to 30% higher than planned, forcing enterprises to expand FinOps practices. Chinese models offering frontier performance at a fraction of the price present an obvious way to manage those costs.

But cost is only part of the equation. Data sovereignty^✦ concerns, regulatory compliance requirements, and the geopolitical complexities of US-China tech competition all factor into procurement decisions. As we explored in our analysis of Chinese AI's market expansion, these dynamics are reshaping enterprise decision-making across the region.

Open Source as Strategic Advantage

A critical factor in China's token dominance is the near-unanimous embrace of open source by Chinese AI firms. Unlike American competitors, who largely keep their frontier models proprietary, Chinese companies including MiniMax, Alibaba, and DeepSeek have released model weights openly.

This earns developer goodwill, drives adoption, and makes it harder for any single government to restrict access. MiniMax's monthly active users grew from 3.1 million in 2023 to 27.6 million by September 2025, a trajectory powered by open access and aggressive pricing. The broader implications of this approach are evident in how Chinese startups are challenging Silicon Valley giants through fundamentally different distribution models.

Is MiniMax M2.5 really as good as Claude Opus 4.6?

On the SWE-Bench coding benchmark, MiniMax M2.5 scores 80.2% compared to Claude Opus 4.6's 80.8%, a gap of just 0.6 percentage points. For many enterprise coding and automation tasks, the performance difference is negligible. However, benchmarks do not capture every real-world use case, and differences in reasoning, safety features, and multilingual performance may still favour specific models for specific applications.

Why are Chinese AI models so much cheaper?

Chinese AI companies benefit from lower operational costs, aggressive pricing strategies aimed at market share capture, and open-source distribution models that reduce support overhead. MiniMax has also optimised its inference^✦ infrastructure for cost efficiency, passing savings directly to users in a bid to establish market dominance.

What does this mean for data security and compliance?

Enterprise buyers must evaluate Chinese AI models against their specific regulatory requirements, data residency rules, and risk tolerance. While many Chinese models offer on-premises deployment options, organisations handling sensitive data should conduct thorough due diligence on data flows, model training sources, and compliance frameworks.

Can Western AI companies compete on price?

Western AI companies face higher operational costs and different investor expectations around profitability timelines. However, some are beginning to respond with more aggressive pricing tiers and efficiency improvements. The competitive pressure from Chinese models may accelerate these efforts across the industry.

How sustainable are these low prices?

Current Chinese pricing appears to be part of a market-capture strategy rather than purely cost-driven. Whether these prices remain sustainable long-term will depend on achieving scale, operational efficiency gains, and the companies' ability to monetise their growing user bases through premium services or enterprise contracts.