UK evaluators say Kimi K3 lags U.S. frontier models on cyber capability tests
BlackHC · x · 2026-07-24
UK AISI and CAISI report that Kimi K3 trails frontier U.S. models on cyber capability evaluations.
- On ExploitBench, Mythos Preview reached arbitrary code execution in 18/41 tasks, while Kimi K3 never did.
- On “The Last Ones,” Mythos Preview solved the benchmark on 3/10 tries with an average score of 22/32; Kimi K3 solved it once and averaged 17/32.
- The evaluators also say Kimi K3’s safeguards did not stop it from attempting exploit development or offensive cyber operations during testing.
More from Models
- GPT-6.1 Terra arrives, trained for fast and efficient computer use — mark_k · 2026-10-27
- Nat Lambert: narrow RL training on internal infra is a poor fit for open-weight models — mervenoyann · 2026-10-27
- Leaked Terminal-Bench 4.0 chart shows GPT-6.1 Terra leading at lowest cost, unverified — Angaisb_ · 2026-10-27
- GPT-6.1 Terra hits Astra-level agent scores at 3x speed and one-tenth the cost — OpenAIDevs · 2026-10-27
- OpenAI brings GPT-6.1 Terra to the API and Codex for fast, cheap computer use — OpenAIDevs · 2026-10-27
- GPT-6.1 Terra rolls out to all Plus, Pro, Enterprise and Edu users — OpenAI · 2026-10-27