Beijing.- The Chinese startup Moonshot AI recently unveiled Kimi K3 as the largest open-source artificial intelligence (AI) model to date, with which the company claims to outperform systems from OpenAI and Anthropic in several tests, in a new episode of the technological struggle between Beijing and Washington.
Moonshot AI noted that Kimi K3 reaches 2.8 trillion parameters—the measure that indicates the complexity of an AI system during its training—a figure above other Chinese open models such as DeepSeek V4 Pro (1.6 trillion) or Zhipu AI’s GLM-5 series (744 billion).
The company noted that the model is geared towards long-range programming tasks, intellectual work, and reasoning, and is already available through the Kimi chatbot, the Kimi Work app, the Kimi Code tool, and its API, with the full release of its weights scheduled for July 27.
It approaches Claude or ChatGPT at a lower price
Moonshot acknowledged that K3’s overall performance “still lags behind the most powerful proprietary models,” Claude Fable 5 and GPT-5.6 Sol, but asserted that it consistently outperformed the rest of the models analyzed.
In an analysis conducted by the AI model evaluation platform Arena.AI, it took the top spot in six out of seven categories—branding and marketing, reference-based design, data and analytics, consumer products, simulations, and content creation tools—placing second only in the video games category, behind Fable 5.
More Read
We recommend reading:SpaceX suffers its second postponement in a week after aborting a Starlink flight
The consulting firm Artificial Analysis placed K3 at a level comparable to Claude Opus 4.8 and GPT-5.5, considered the second-best models from Anthropic and OpenAI, respectively, although behind Claude Fable 5 and GPT-5.6 Sol in overall performance.
The firm highlighted that the performance improvement of K3 comes with a higher price: 15 dollars per million tokens generated, which places the average cost per task at 0.95 dollars.
More Read
Although it remains more expensive than other open-weight Chinese models, it sits below the $1.04 of GPT-5.6 Sol and at almost half the $1.80 of Claude Opus 4.8.
Beijing-Washington technological struggle
The announcement comes in a context of growing rivalry among major Chinese developers—Moonshot, Zhipu, DeepSeek, and Alibaba, among others—who are betting on openness, low prices, and rapid commercialization to close the gap with the United States, and just a month after Zhipu introduced its GLM-5.2 model.
In addition to US models, they face Washington’s restrictions on the export of advanced chips and the machinery to manufacture them, although Bank of America analysts, cited by CNBC, noted that K3 demonstrates that Chinese firms continue to make progress despite those limitations.
The launch coincided with the opening in Shanghai of the World Artificial Intelligence Conference, an area set by Chinese authorities as strategic, where Xi Jinping advocated for open source so that “all sectors and companies can benefit from AI.”
