An Unexpected Breakthrough: How Does the Launch of a Cutting-Edge Neural Network in China Threaten Anthropic and OpenAI?

Developments from Chinese artificial intelligence labs are rapidly catching up to their American counterparts in terms of performance, according to The Wall Street Journal (WSJ). For example, on July 17, the Chinese AI startup Moonshot AI— whose investors include the venture capital firm HSG and the tech giant Alibaba—unveiled what it claims is the world’s largest open-source AI model, Kimi K3, according to Reuters. Its creators claim that its performance is approaching that of Anthropic’s cutting-edge AI model, Fable.

In most tests involving programming and working with AI agents, Kimi K3 outperformed Anthropic’s Claude Opus 4.8 AI model, released in late May, and OpenAI’s GPT 5.5, launched in late April, according to the developers at Moonshot AI (although, they say, the model still lags behind leading U.S. proprietary models, such as Anthropic’s Fable 5 and OpenAI’s GPT-5.6 Sol). Nevertheless, Kimi K3 delivers performance approaching that of Anthropic’s cutting-edge Fable model, according to the AI startup.

Overall, according to Moonshot, Kimi K3 is the first open-source neural network to approach the 3 trillion-parameter mark. This number indicates the size of the model’s neural network, explains the Financial Times—generally, the more parameters there are, the greater the model’s capabilities. Not all AI labs disclose this data. By way of comparison, independent researchers estimate that Claude Opus 4.8, released in May, contains more than 1.5 trillion parameters.

Kimi K3 is designed for complex logical reasoning, as well as long-form programming and information processing, Reuters explains. The model has a context window of 1 million tokens, which allows it to process and store significantly more data within a single query than previous generations.

Moonshot plans to fully open-source Kimi K3 by the end of this month, after which users will be able to freely download and adapt it, the WSJ notes.

Analysts at Jefferies called the Moonshot release an “unexpected breakthrough” and said they expect the AI company to continue driving innovation in the field of AI. China’s rapid progress in the field of artificial intelligence is partly driven by Beijing’s desire for self-sufficiency amid competition with the U.S. for technological supremacy, explains the WSJ.

The release of K3 could challenge the industry consensus that Chinese AI models lag behind their American counterparts by eight to twelve months in terms of performance, the FT notes. At a conference on Wednesday, Tarun Chhabra, head of national security policy at Anthropic, acknowledged that the U.S. lead over Chinese labs in the field of artificial intelligence models currently stands at “roughly six to nine months” (quoted in the WSJ).

The open-source nature of the Chinese model could also pose problems for U.S. AI labs, such as Anthropic and OpenAI, whose expensive, cutting-edge neural networks remain proprietary, the Financial Times adds. Open-source code provides users with virtually free access to the models and accelerates their adoption among developers and companies, allowing for fine-tuning to specific needs.

It is precisely because many Chinese neural networks are open-source and, as a rule, cheaper to operate than their Western counterparts that AI models from China have become widely adopted around the world. The FT notes that, as a result, companies in Silicon Valley and Europe are increasingly switching to cheaper Chinese AI models.

However, in late June, Anthropic accused several Chinese labs, including Moonshot and Z.AI, of using American models to train their own.

Moonshot's announcement sent the shares of the AI startup's Chinese competitors tumbling: Zhipu (Z.AI) and Minimax shares fell by nearly 28% and 16%, respectively, on the Hong Kong Stock Exchange.

This spring, Tech in Asia, as well as Bloomberg and the WSJ, reported—citing sources—that Moonshot had begun preparations for an initial public offering in Hong Kong. According to WSJ sources familiar with the situation, the Chinese AI startup is valued at $31.5 billion as part of its ongoing funding round.

Another Chinese AI lab, DeepSeek, is also preparing for an IPO on the Shanghai Stock Exchange—it is valued at $51.82 billion, according to Reuters. In May, following another round of fundraising, Anthropic reached a valuation of $965 billion. OpenAI’s latest valuation stood at $852 billion.

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