Gap in Long-Range Cybersecurity Capabilities Narrows
rohanpaul_ai · x · 2026-07-18
This post outlines evaluation results for long-horizon cybersecurity capabilities: on AISI's 32-step "The Last Ones" cyber range, the gap between the latest open-weight models and closed-source frontier models has narrowed from about 6–10 months to roughly 4–7 months.
The chart provides two insights: on the long-horizon cyber range, GLM-5.2 performs similarly to closed-source models released about 4 months ago; on AISI's 32-step task, GPT-5.6 Sol completed all 32 steps 7 out of 10 times under a 100M token budget. The author emphasizes that as inference tokens increase, model capabilities continue to rise, meaning attack/defense capabilities can be significantly enhanced through more runtime compute without requiring retraining.
Related event: AISI says open models narrow the cyber-range gap(6 posts)→
More from Models
- Kimi K3 is praised for stronger English, frontend arena #1, and better handling of nuanced prompts — EXM7777 · 2026-07-22
- Gary Marcus says LLM math skills are like knowing only a car’s engine size — GaryMarcus · 2026-07-22
- OpenAI’s Codex + GPT-5.6 Sol hits 99% recall in Project APE verification tests — soumitrashukla9 · 2026-07-22
- OpenAI-linked paper says capability RL can make models more reward-seeking — MariusHobbhahn · 2026-07-22
- Macaron V1 adds LoRA RL on GLM 5.2 and claims SOTA benchmark gains — Xianbao_QIAN · 2026-07-22
- OpenAI rolls out voice in GPT-Live, but the UI obscures search and reasoning — Graham_dePenros · 2026-07-22