China's AI sector narrows gap with US through efficiency and open-source models

Artificial Intelligence Large Language Models China-US tech rivalry AI model distillation Open-source AI

Chinese artificial intelligence companies are rapidly closing the performance gap with leading US developers through strategies of efficiency, open-source accessibility, and government support. Alibaba has unveiled Qwen3. 8-Max, its most capable model to date, which reportedly rivals top systems from OpenAI and Anthropic in text analysis and video processing.

To mitigate the impact of US chip sanctions, Chinese firms like DeepSeek and Moonshot AI are focusing on high-efficiency architectures. DeepSeek utilizes techniques such as Multi-head Latent Attention (MLA) and Mixture-of-Experts (MoE) to reduce computational requirements. This focus on efficiency has allowed Chinese models to offer significantly lower API pricing compared to US counterparts; for instance, DeepSeek's V4 Flash model is priced at a fraction of the cost of similar tasks performed by Anthropic.

Furthermore, the Chinese ecosystem benefits from a robust talent network and state-level support, including electricity subsidies for data centers using domestic hardware. While researchers have identified potential vulnerabilities that allow for the extraction of hidden reasoning traces from advanced models—raising questions about whether some Chinese models were trained via 'distillation' of US-made systems—the overall trend indicates that the technological gap between the two nations has narrowed to a matter of months.

Entities

Alibaba · Anthropic · DeepSeek · Moonshot AI · OpenAI

Claims

What the coverage asserts, and how many sources carry each claim.

[○ 1 SOURCE] Moonshot's Kimi K3 model shows performance almost equivalent to Anthropic's latest models. www.sisajournal.com
[○ 1 SOURCE] Researchers discovered a method to extract hidden reasoning traces from advanced AI models. wired.jp
[○ 1 SOURCE] DeepSeek's V4 Flash model charges $0.22 per 1 million input tokens and $0.66 per 1 million output tokens. www.sisajournal.com
[○ 1 SOURCE] China provides electricity subsidies for data centers that use domestic chips. www.sisajournal.com
[○ 1 SOURCE] The gap between China's best AI models and the US's most advanced models has narrowed to months. www.ithome.com
[● 2 SOURCES] Chinese AI models focus on efficiency to mitigate the impact of US chip sanctions. www.sisajournal.com · www.ithome.com
[○ 1 SOURCE] Anthropic charges $10 per 1 million input tokens and $50 per 1 million output tokens for similar tasks. www.sisajournal.com
[○ 1 SOURCE] Moonshot AI's Kimi K3 model produced reasoning traces similar to Claude Opus and GPT-5. wired.jp
[○ 1 SOURCE] Alibaba's Qwen3. 8-Max model shows performance comparable to leading US systems from Anthropic and OpenAI. bithub.si