“Moonshot AI Kimi” app. AFP-Yonhap News
A Chinese artificial intelligence (AI) startup has once again put US Big Tech on edge. After shocking the industry last year with its low-cost AI model “DeepSeek,” it has now unveiled “Kimi K3,” which delivers performance approaching that of top-tier US models.
According to industry sources on the 19th, Chinese AI startup Moonshot AI recently unveiled its next-generation AI model “Kimi K3.” Released ahead of the opening of the World Artificial Intelligence Conference (WAIC 2026) in Shanghai, China, Kimi K3 features native multimodal capabilities that process images and text simultaneously, along with an ultra-large context window that handles up to 1 million tokens at once.
Moonshot AI introduced Kimi K3 as a model specialized in long-form coding, complex reasoning, and expert knowledge tasks, describing it as the world’s first “3-trillion-parameter-class open-weight AI model.”
China’s Moonshot AI model “Kimi K3.” Captured from Kimi website
Approaching Top US Models: Competing on Performance Instead of Low Prices
The most notable feature of Kimi K3 is that it moves away from the “ultra-low-price strategy” that Chinese AI companies have championed, shifting toward performance competition.
Kimi K3 is equipped with approximately 2.8 trillion parameters. This far exceeds existing Chinese open models such as DeepSeek V4 Pro (1.6 trillion) and Zhipu AI’s GLM-5 series (740 billion). Parameters are internal variables that AI learns during training, and generally, the more parameters, the higher the likelihood of processing more complex information.
Its performance also approached top-tier US models. AI performance evaluator Artificial Analysis rated Kimi K3’s intelligence index at 57 points. While it fell somewhat short of Anthropic’s Claude Fable 5 (60 points), OpenAI’s GPT-5.6 Sol (59 points), and GPT-5.6 Sol X-High (58 points), it surpassed xAI’s Grok 4.5 and major models from Meta and Google.
Its coding ability also received high marks. It ranked first in Arena AI’s web interface development evaluation and second, behind Claude Fable 5, in Vals AI’s comprehensive evaluation, scoring higher than GPT-5.6 Sol.
Its price is higher than that of Chinese competitors. API fees are around $3 per 1 million input tokens and $15 per 1 million output tokens. This is similar to GPT-5.6 Sol but more expensive than other Chinese open models such as DeepSeek V4 Pro or the GLM-5 series. The industry has analyzed that Moonshot AI has shifted its direction toward a strategy targeting the enterprise AI market rather than simple price competition.
Shaking New York, Tensing KOSPI: Samsung and SK hynix in Play
US markets responded immediately after Kimi K3’s release. On the 17th, the Nasdaq index fell 1.40%, the S&P 500 index fell 1.01%, and the Dow Jones Industrial Average fell 0.77%. Nvidia dropped 2.2%, Applied Materials 5.6%, Intel 2%, and SanDisk 4%.
The market focused on two points. First is the possibility that as near-free open AI models rapidly advance, profitability could weaken for OpenAI, Anthropic, and Google, which sell subscription-based closed models. At the same time, concerns have grown that the timeline for recovering AI infrastructure investments could be delayed longer than expected.
SK Group Chairman Tae-won Choi also said at the Korea Chamber of Commerce and Industry’s summer forum on the 17th, “Memory demand will increase as time goes on.” He added, “Looking at the AI field alone, next year’s demand is expected to increase by at least 60 to 100% compared to this year. Rather than being shaken by short-term stock prices, we should look at the long-term trend.”
Samsung Electronics, SK Hynix. Yonhap News
The KOSPI market is also being mentioned as potentially affected. Samsung Electronics (005930.KS) and SK hynix (000660.KS) have been evaluated as the biggest beneficiaries of expanded AI investment, and if controversy over AI investment efficiency resurfaces, the “peak theory” for the memory semiconductor sector could reignite.
Of particular interest is the fact that Moonshot AI did not disclose which chips it used amid US regulations on high-performance AI chip exports. The industry has raised the possibility that Chinese-made AI semiconductors, such as Huawei Ascend, and Chinese-made high-bandwidth memory (HBM) accounted for a substantial share. According to analysts, if a Chinese AI model proves its competitiveness using only domestically produced semiconductors, this could become a burden factor for Samsung Electronics and SK hynix.
However, some forecast that the market impact will be limited. This is because when DeepSeek R1 was unveiled last year, US AI-related stocks also plunged but subsequently recovered quickly.
Patrick Moorhead, CEO of market research firm Moor Insights & Strategy, told the SCMP, “This reaction is an overreaction similar to the DeepSeek episode,” adding, “Kimi K3 will further accelerate the growth of the AI inference market.”