Moonshot AI launched its Kimi K3 model last week, and demand overwhelmed its servers within hours. The Chinese startup suspended new subscriptions after the open-weight model attracted millions of users in a single weekend.

For free. That is the key detail. Unlike OpenAI’s GPT-4, Anthropic’s Claude, and other leading US models, Kimi K3 is available to download and run on your own hardware. Users can modify it, retrain it, and deploy it without paying anyone.

What Kimi K3 Can Do

The model ranks at the top of major coding benchmarks. It outperforms most rivals on reasoning tasks and matches or exceeds the latest iterations of GPT and Claude on general benchmarks. Kimi K3 is not state-of-the-art—Claude Fable 5 still leads most tests—but for zero cost, it is a serious competitor.

The model is based on 2.8 trillion parameters, making it massive but not the largest. Speed appears to be a strength. Users report faster inference than comparable models, which matters when you are running thousands of queries per day.

The Market Shock

Taiwan Semiconductor Manufacturing Company fell 7% on the day Kimi K3 launched. Investors are pricing in a world where frontier AI models are cheap and abundant, not scarce and expensive. That logic favors Kimi K3.

Moonshot raised $2 billion at a $20+ billion valuation in May. The company is based in Beijing and is one of China’s leading model builders. Kimi K3 is proof the US does not hold a monopoly on advanced AI.

What Happens Next

Moonshot will likely re-enable subscriptions once capacity scales. The company has shown it can build competitive models fast. The question now is whether it can monetize them without losing users to open-weight alternatives. Kimi K3 has already set a new expectation: frontier AI should be free or very cheap.

OpenAI, Anthropic, and Google are all watching this closely. The economics of proprietary AI just shifted.