The Kimi K3 model demonstrates performance parity with Fable and 5.6 Sol in coding and computer use, while emerging as the top model for cybersecurity tasks due to competitor refusals. This performance significantly narrows the capability gap for open-weight models against proprietary frontier models to 1.25-3.5 months in core AI use cases. However, early evaluations suggest Kimi K3 lags in other domains like math and genetics.
The capability gap for Chinese open-weight models in economically critical tasks like coding and cybersecurity is shrinking, intensifying competitive pressure on US frontier labs.