KM4BI KM4BI

China's K3 Model Reveals the Problem With Open Weights

Kimi K3 shows that open-weight models can be powerful while still being expensive, compute-heavy, and behind closed frontier labs.

Moonshot’s Kimi K3 is presented as a major open-weight release, but Nate B Jones argues that it breaks a common assumption: open does not automatically mean cheap, small, or easy to run. The model appears strong in coding, yet its full-performance footprint is closer to corporate infrastructure than home compute.

The economic issue matters. K3 is not just expensive because of the hardware needed to serve it; it is also expensive because its token pricing is relatively high and because it uses more tokens to reach an answer than the most efficient closed frontier models. That weakens the idea that Chinese open models are necessarily winning through serving efficiency.

Jones still sees real value. K3 can help with use cases that closed models may restrict, including fine-tuning and certain kinds of software analysis or replacement work. Its value is therefore capability and flexibility, not a simple promise of lower cost.

The video also treats K3 as a warning sign for cybersecurity. As open models become stronger, they can be used by bad actors for cyber operations, social engineering, and software attacks. The practical response is defense in depth: stronger authentication, software audits, and identity-verification habits for families and teams.

Strategically, Jones expects a more fragmented model future. Governments may become more involved in restricting advanced model distribution, so both individuals and companies should plan around multiple providers, backup models, and a mix of local and cloud access.

Source