Chinese AI Models Dominate Global Usage: Top 4 Spots Held by Domestic Tech

Chinese artificial intelligence large language models have surpassed their American counterparts in weekly API call volume for the sixth consecutive week. This shift marks a significant turning point in the global AI race, highlighting the rapid adoption and technical maturity of domestic Chinese technologies.

According to data from OpenRouter, the total global AI model call volume reached 36.1 trillion tokens last week. This represents a 13.5% increase from the previous week, continuing a seven-week upward trend in overall market activity.

Key Takeaways

• Market Leadership: Chinese AI models achieved 14.19 trillion token calls last week, surpassing US models which recorded only 3.2 trillion.
• Growth Divergence: Chinese model usage grew by 27.49% week-over-week, while US model usage dropped by 24.53%.
• Top Rankings: The top four most-called models globally are all Chinese, led by DeepSeek-V4-Flash.
• New Entrants: MiniMaxM3 entered the top three immediately upon release, showcasing strong initial market traction.
• Total Volume: Global AI interaction volume hit 36.1 trillion tokens, indicating sustained high demand.
• Competitive Landscape: The gap between Western and Eastern AI adoption is narrowing rapidly in specific developer segments.

DeepSeek Leads the Charge

DeepSeek-V4-Flash has maintained its position as the world's most-called AI model for three straight weeks. The model processed 3.69 trillion tokens last week alone. This figure represents a 19% increase compared to the prior week, demonstrating robust user retention and growing trust in its capabilities.

The success of DeepSeek suggests that developers are prioritizing cost-efficiency and performance balance. Unlike earlier models that focused solely on raw parameter counts, DeepSeek appears to optimize for inference speed and economic viability. This approach resonates strongly with enterprise clients managing tight budgets.

Tencent Hy3 Preview Holds Strong

Tencent’s Hy3 preview model secured the second spot globally. It handled 2.94 trillion tokens during the same period. Although this shows a slight 3% decline from the previous week, it remains a dominant force in the Asian market.

Tencent’s integration of Hy3 into its vast ecosystem likely drives consistent usage. The model benefits from immediate access to millions of users through WeChat and other Tencent platforms. This distribution advantage allows for rapid iteration and real-world testing at a scale few competitors can match.

MiniMaxM3 Shatters Expectations

MiniMaxM3 made a stunning debut by jumping straight into the top three. The model processed 2.5 trillion tokens in its first week of availability. Such an immediate surge indicates strong pre-release anticipation and effective marketing strategies.

Official documentation highlights MiniMaxM3 as the first domestic model to combine cutting-edge coding abilities with a 1 million token context window. It also features native multimodal capabilities. These specifications place it in direct competition with leading Western models like GPT-4o and Claude 3.5.

Technical Superiority in Context

The 1M context window is a critical differentiator for enterprise applications. Developers can now process entire codebases or extensive legal documents in a single prompt. This reduces the need for complex retrieval-augmented generation (RAG) systems in many scenarios.

Native multimodal support further enhances its utility. Users can analyze images, charts, and text simultaneously without switching models. This seamless integration improves workflow efficiency for developers and data analysts alike.

US Models Face Adoption Headwinds

While Chinese models surged, US-based models experienced a notable decline. The total weekly call volume for American models fell by 24.53% to 3.2 trillion tokens. This sharp drop contrasts with the 27.49% growth seen in China.

Several factors may contribute to this downturn. Higher API pricing from major US providers could be driving developers toward more affordable alternatives. Additionally, rate limits and capacity constraints might be pushing users to diversify their model dependencies.

Cost Efficiency Drives Migration

Developers are increasingly sensitive to operational costs. Chinese models often offer competitive pricing structures that appeal to startups and mid-sized enterprises. This economic incentive accelerates migration away from premium-priced Western alternatives.

The open-source nature of many Chinese models also facilitates local deployment. Companies concerned about data privacy can host these models on-premise. This flexibility is less common with closed-source US APIs, creating a strategic advantage for Chinese tech firms.

Industry Context and Broader Implications

This data reflects a broader trend in the global AI landscape. The monopoly of Silicon Valley on large language model innovation is ending. Regional players are developing specialized models that cater to local languages and regulatory environments.

For Western businesses, this signals a need for strategic reassessment. Relying exclusively on US-based APIs may no longer be the most efficient or cost-effective strategy. Diversifying model providers ensures resilience against price hikes and service disruptions.

What This Means for Developers

• Cost Savings: Switching to top-tier Chinese models can reduce inference costs significantly.
• Performance Parity: Models like MiniMaxM3 offer context windows and coding skills comparable to GPT-4.
• Data Sovereignty: Local hosting options provide better control over sensitive information.
• Innovation Speed: Rapid iteration cycles in China lead to faster feature releases.

Looking Ahead

The next few months will be crucial for determining if this trend is temporary or structural. US providers must respond with competitive pricing and enhanced features to retain market share. Failure to do so could result in a permanent loss of dominance in the developer community.

We expect to see increased collaboration between Eastern and Western tech firms. Joint ventures and licensing agreements may become more common as the technology matures. This cross-pollination will accelerate global AI advancement.

Gogo's Take

• 🔥 Why This Matters: The dominance of Chinese models proves that high-quality AI is no longer exclusive to Silicon Valley. Businesses can achieve similar performance at lower costs, democratizing access to advanced AI tools globally.
• ⚠️ Limitations & Risks: Reliance on foreign models introduces geopolitical risks and potential supply chain disruptions. Data privacy concerns remain paramount when using APIs hosted outside one's jurisdiction.
• 💡 Actionable Advice: Developers should benchmark MiniMaxM3 and DeepSeek-V4-Flash against current US models. Test them on specific tasks like coding and long-context analysis to identify immediate cost-saving opportunities.