China's New AI Breakthroughs Could Leave America Further Behind

China is making new waves in the world of artificial intelligence (AI). Several of the country’s technology companies have launched their latest advanced AI models.

Alibaba introduced Qwen3.8-Max, its largest and most sophisticated model to date, while DeepSeek presented a model with operational costs claimed to be over 100 times cheaper than Anthropic’s Claude Fable 5.

Alibaba’s Qwen3.8-Max immediately boosted the company’s stock by 7% in Hong Kong trading. Simultaneously, research firm Artificial Analysis named DeepSeek’s V4-Flash as one of the world’s cheapest AI models.

Both models reinforce Chinese AI firms’ commitment to an open-weight approach, which allows developers to download and customise the core parameters of the AI system. This stands in contrast to the closed models used by OpenAI, Anthropic, and Google.

Lian Jye Su, Chief Analyst at Omdia, noted that Chinese AI companies have identified a crucial market. “Many business workflows do not require the industry’s best AI model,” he said, adding that businesses need models that are sufficiently good, affordable, transparent, and easily accessible, with open-weight models helping to meet these needs.

Qwen3.8-Max immediately ranked highly on various leaderboards measuring AI capabilities. The model features 2.4 trillion parameters—the numerical configurations AI learns from data to recognise patterns, generate answers, and complete tasks. It became the highest-ranked Chinese AI model for text on the Arena.AI platform, though it still trails Anthropic’s Claude Fable 5 and several Opus variants.

Capable of processing text, images, and video, Qwen3.8-Max can handle up to 1 million tokens at once, enabling it to process lengthy legal documents, large software codebases, and hundreds of pages of documents simultaneously. Alibaba stated the model will be released next week and successfully completed a software engineering project in 16 days.

The model uses a mixture-of-experts architecture, which divides work among specialised AI components so the entire model is not active for every request. This means only about 95 billion parameters are used per processing task, reducing operational costs and response times.

Meanwhile, DeepSeek launched V4-Flash on Friday. According to Artificial Analysis testing, it is the world’s lowest-cost prominent AI model. The Chinese AI startup previously captured global attention in early 2025 with its R1 and V3 models, triggering a global tech stock sell-off and raising questions about the scale of US companies’ AI spending.

DeepSeek priced V4-Flash at US$0.14 per million input tokens and US$0.28 per million output tokens. Artificial Analysis estimated the average cost per test at just US$0.03, compared to US$0.86 for Kimi K3, US$1.86 for OpenAI’s GPT-5.6 Sol, and US$3.15 for Claude Fable 5. This comparison is considered more reflective of real value as it accounts for the amount of data the model must process and generate to complete a task.

With a combination of massive capacity and extremely low operational costs, Chinese AI companies are further strengthening their position in the global competition.