Chinese AI startup Moonshot AI on July 16 (local time) released its open-weight model Kimi K3, which has become a focus of attention in Silicon Valley. U.S. media outlet Business Insider (BI) on July 17 (local time) highlighted why Kimi K3 is drawing attention and the reactions of industry figures.

Kimi K3 has 2.8 trillion parameters, making it the largest open-weight model released so far. It can process documents running to hundreds of pages at once, showing strength in analyzing long documents or large codebases. On the front-end coding leaderboard of AI evaluation platform Arena, it took first place, beating Anthropic's Claude Fable 5.

Price competitiveness is also a key weapon. Kimi K3's API costs $3 per 1 million input tokens and $15 per 1 million output tokens. That is less than half the level of OpenAI's GPT-5.6 Sol at $5 and $30, and Anthropic's Claude Fable 5 at $10 and $50. Moonshot AI has received investment from Alibaba and Tencent and is valued at about $31.5 billion, and it plans to release model weights as open source on July 27. The announcement came a day before the opening of the 2026 World Artificial Intelligence Conference (WAIC) in Shanghai on July 17.

■ Open source shakes the standing of closed models... but some urge caution, asking, "Can it be trusted?"

Vercel CEO Guillermo Rauch (기예르모 라우치) called it the first time in this comprehensive web engineering benchmark that an open model outperformed all closed models, but added that benchmarks do not tell the whole story. Box CEO Aaron Levie (애런 레비) said it was "really surprising" for an open model to perform at this level and called it a "big win" for companies building with AI. He predicted that lower-cost frontier-level intelligence would lead directly to expanded corporate use of AI.

Venture capitalist Jason Calacanis (제이슨 칼라카니스) said, "Things are happening," and claimed AI progress has been faster in the past 30 days than in the past 30 months. He also offered a view that 2026 would be the year of artificial general intelligence (AGI) and that superintelligence would emerge in 2027 to 2028. Professor Russ Salakhutdinov (러스 살라후트디노프), who was the Carnegie Mellon University (CMU) PhD adviser to Moonshot founder Yang Zhilin (양즈린), congratulated his student’s achievement, calling it a "big win for the open-source community."

Wharton School professor Ethan Mollick (이선 몰릭) said Kimi K3 is "the closest to the frontier" among models released so far, but advised against judging it by benchmark scores alone. He said that after assigning Kimi K3 a statistical audit of his past academic paper, he found multiple errors, including misapplication of statistical methods. Former Meta product manager Xiaoyin Qu (샤오인 취) asked whether the United States can maintain a technological edge when only OpenAI and Anthropic show comparable performance.

Political reactions were sharper. David Sacks (데이비드 색스), an AI adviser to the Trump administration, said it was the first time a Chinese model had taken first place in front-end coding and called it "worrisome." He criticised what he described as data centre regulation, a proliferation of state-level rules and demands for federal pre-approval, saying, "This is how the United States falls behind in the AI race." AI industry critic Gary Marcus (게리 마커스) cited a Goldman Sachs analysis that U.S. cloud companies' capital investment would reach about $1 trillion in 2027, eight times China’s, and argued that "Congress should investigate."