Study says Kimi K3 scores 32.2% on cyberattack ability vs 76.2% for U.S. models
pstAsiatech · x · 2026-07-25
A study cited by SCMP says Kimi K3 trails unnamed U.S. models on cyberattack capability.
The reported numbers are stark: Kimi K3 scores 32.2% overall, while the top U.S. models average 76.2%. The post also asks whether Moonshot has under-optimized for coding, implying the model may not be tuned strongly for offensive security or code-heavy tasks.
Related event: Kimi K3 Cybersecurity Eval Sparks Debate: Scores 32.2%(5 posts)→
More from Models
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27