UK, US Tests Find China's Kimi K3 AI Lags Top American Models ...

The assessment focused on cyber capabilities, including tasks such as vulnerability discovery and exploitation. The test was designed to compare Kimi K3 against leading American models' abilities in offensive and defensive cybersecurity scenarios.

The findings indicate that Kimi K3 lags behind in both speed and success rate across a range of cyber-specific benchmarks. Moonshot AI has not yet publicly responded to the assessment. The evaluation is described as 'preliminary', suggesting further detailed analysis may follow. Both AISI and CAISI were established to evaluate frontier AI systems for risks including those in cybersecurity.

The report did not include specific scores or rankings, but stated the gap is 'significant' in the tested areas.