Chinese AI developers have created a "death zone" for Anthropic and OpenAI

The increase in the number of new AI model releases launched by Chinese AI labs is narrowing the gap between them and their American counterparts and creating what researchers at the independent analytics firm Artificial Analysis have called a “death zone” for AI developers worldwide. Beyond this zone, advanced and relatively inexpensive AI models from China could leave behind anyone who lacks innovative technologies or does not set market prices, Bloomberg reports.

Launched in April by Chinese AI developer DeepSeek, the large language model V4 Flash "has transformed the economics of agent-based workloads," says Dermot McGrath, founder of ZenGen Labs, a Shanghai-based consulting firm specializing in startups. Tests showed that V4-Flash is about 105 times cheaper than Anthropic’s cutting-edge AI model, Claude Fable 5, according to a conclusion reached by Artificial Analysis, as reported by Reuters. And although V4’s performance, according to independent researchers, still lags behind developments from OpenAI and Anthropic, the gap between them is rapidly narrowing—the DeepSeek model is already comparable to Google’s Gemini 3.6 Flash, according to an analysis by Artificial Analysis.

In addition, Alibaba unveiled its largest AI model, Qwen3.8-Max, this week, which, according to its creators, matches or surpasses Anthropic’s flagship Fable 5 model. And a few weeks earlier, another Chinese AI lab, Moonshot, released the Kimi K3 neural network, whose performance is comparable to the most expensive American counterparts, according to Bloomberg. Both of these models are priced slightly higher than DeepSeek’s offering, but are still cheaper than leading American counterparts from OpenAI and Anthropic.

Now, Bloomberg continues, in order to win over customers, AI companies must either offer a more competitive price than DeepSeek or outperform Chinese AI models in terms of capabilities—something Artificial Analysis refers to as the “death zone” for competitors of Chinese AI labs.

“The most significant change since January 2025 is that China’s progress no longer appears to be the breakthrough of a single company,” noted Po Zhao, a tech analyst and author of the Hello China Tech newsletter, while commenting on the situation. “DeepSeek’s first breakthrough [a year ago] seemed exceptional. The latest data, however, indicates that China now has a system for creating AI models that are close to global standards,” he added.

McGrath told Bloomberg that to carry out his day-to-day work as an architect—to draw up an action plan—he still uses Claude from Anthropic, but then delegates the execution of those tasks to the DeepSeek model, which deploys specialized AI agents to complete the task.

“A few months ago, I wouldn’t have done that,” the analyst remarked. “Chinese models weren’t as reliable when it came to invoking tools or executing lengthy agent workflows,” he emphasized.

This article is being updated

This article was AI-translated and verified by a human editor