Kimi K3 Beats Cyber Benchmark
China's new 2.8T model just matched US frontier AI in cybersecurity. Is the US falling behind?
What the video says
An open-weight model just uncovered 23 out of 26 vulnerabilities on Aikido Security's private benchmark. This is Moonshot AI's new 2.8 trillion parameter model, KIMI-K3.
It matches US frontier capabilities in critical cybersecurity tasks. Analysts suggest US labs are falling behind in defensive analysis because their strict safety guardrails and API-only access limit their practical utility.
This pivot-worthy Chinese release lands alongside Black Forest Labs' flagship release of FLUX 3 Video. Two years after hinting at video, the lab delivered native audio generation across all modalities.
The new model supports up to 3-minute one-pass generation and agentic clip chaining. But as open-weight power shifts, US Treasury officials are already considering sanctions to block PRC distillation attacks.
