Kimi K3 Cybersecurity Eval: Improved Long-Chain Execution but Overall Weak

Kimi K3 recently scored 32% on ExploitBench's cybersecurity eval with zero ACEs. Despite criticism that its cyber capability remains weak, the model demonstrated the ability to execute complete, long-chain tasks when compared to other US and Chinese models.

2026-07-24 ~ 2026-07-24 · 2 related posts

• Kimi K3 scores 32% on ExploitBench and reaches 0 of 41 ACE cases — xeophon · 2026-07-24
• Kimi K3’s cyber eval still looks weak, but it can finish a full long-horizon chain — dyn___ · 2026-07-24