"Like DeepSeek a Year Ago"... China's Kimi Bursts Onto the Scene, Shaking Up the Global AI Landscape

Kimi K3 Ranks Fourth After Anthropic and OpenAI in Performance
Chinese AI Shifts From Ultra-Low Prices to Full-Fledged Performance Competition

Chinese artificial intelligence (AI) startup Moonshot AI is shaking up the global AI landscape, traditionally dominated by American companies, by unveiling its new model, Kimi K3. Unlike previous models that competed mainly on price, Kimi K3 demonstrates a threatening presence by delivering performance comparable to cutting-edge American models.

Recently released by Moonshot, Kimi K3 is an AI model boasting 2.8 trillion parameters. Parameters are the weights AI adjusts during training, with a higher number indicating the ability to process more complex information and pointing to the model's overall scale. Thanks to its multimodal understanding and a context window of one million tokens (the amount of text it can refer to when generating a response), Kimi K3 is capable of handling vast amounts of information in a single prompt.

A screen running 'Kimi K3,' the latest AI model unveiled by Chinese AI startup Moonshot AI. Screenshot from the Kimi homepage

Moonshot explained that Kimi K3 is the world’s first open-source model at the three-trillion parameter level, designed with the goal of supporting long-term tasks such as large-scale software repository handling, terminal tool operations, and integrating programming with visual feedback. The company also stated that its image and video understanding capabilities enable users to edit digital content such as interfaces or games.

Closely Chasing America’s State-of-the-Art AI Models

Kimi K3 is recognized for catching up to the latest American AI models in terms of performance. According to AI performance analysis firm Artificial Analysis, Kimi K3 scored 57 points on its intelligence index—ranking fourth among all current AI models. It comes after Anthropic’s Claude Fable (60 points), OpenAI’s GPT-5.6 Sol Max (59 points), and GPT-5.6 Sol XHi (58 points). Kimi K3 also outperforms SpaceXAI’s Grok 4.5 (54 points), Meta’s MuseSpark 1.1 (51 points), and Google’s Gemini 3.5 Flash (50 points).

In ArenaAI’s developer evaluation—which assesses web interface-building capabilities—Kimi K3 ranked first place. In ValsAI’s overall evaluation, it secured second place after Fable, ahead of GPT-5.6 Sol.

Kimi K3 breaks away from the ultra-low price strategy of previous Chinese AI models, instead opting to compete as a high-performance model designed for coding and knowledge work. The API fee for Kimi K3 is 3 dollars per one million tokens for input and 15 dollars for output. While Kimi K3’s prices are higher than those of previous Chinese open-source models, they are relatively inexpensive compared to the leading American models. For example, Claude Opus 4.8 charges 5 dollars for input and 25 dollars for output per one million tokens, while GPT-5.6 Sol charges 5 dollars for input and 30 dollars for output, making Kimi K3 competitively priced.

The debut of Kimi K3 evokes memories of last year’s “DeepSeek Shock,” when Chinese AI company DeepSeek launched its high-performance inference model, DeepSeek R1, causing a stir in the global AI market.

At that time, DeepSeek, reportedly unable to obtain Nvidia’s latest high-performance graphics processing units (GPUs) due to the U.S.-China trade conflict, surprised the world by achieving performance rivaling that of the latest American big tech models. This sparked doubts about whether high-performance AI could be developed without the latest GPUs, leading to a sharp drop in semiconductor stock prices. Following the release of Kimi K3, the Philadelphia Semiconductor Index in the United States also fell for three consecutive trading days, dropping more than 20% from its peak on June 22.

Gap with America’s Most Advanced Models Narrows to 2–3 Months

The unveiling of Kimi K3 is expected to reignite the momentum behind China’s “AI rise,” which had somewhat stalled. Ion Stoica, co-founder of global data firm Databricks and professor of computer engineering at the University of California, Berkeley, told Bloomberg that, “Whereas Chinese models previously lagged behind America’s top models by six to nine months, the current gap appears to have narrowed to just two or three months.”

The Financial Times also noted that the release of Kimi K3 could upend the prevailing belief in the industry that Chinese AI models lag behind their American counterparts by eight to twelve months. In particular, Kimi K3’s open-source approach is seen as a threat to American AI companies like Anthropic and OpenAI, which keep their latest models closed.

The U.S. policy of making AI models into strategic assets has also been unavoidably affected. U.S. President Donald Trump last month signed an executive order requiring American companies to undergo government security reviews before releasing their latest high-performance AI models, and temporarily restricted foreign access to Anthropic’s Mythos and Fable models. In addition, exports of high-performance GPUs like Nvidia’s Blackwell to China remain restricted.

Chinese President Xi Jinping, in his keynote speech at the World Artificial Intelligence Conference (WAIC) held on July 17 (local time), the day Kimi K3 was unveiled, emphasized that, “AI development should not be a solo effort by one country, but a symphony of international cooperation.” Aiming at the U.S. strategy of turning AI models into strategic assets, President Xi added, “We should work together to build a fair and rational system of global AI governance.”

This content was produced with the assistance of AI translation services.