Vals AI's CyberOffense Shows DeepSeek V4.1 Flash Hacking Corporate Networks for Under $3

Vals AI released CyberOffense, a benchmark that gives AI agents up to 48 hours to compromise simulated corporate networks. Open-weight DeepSeek V4.1 Flash 1017 completed 16.3 of 21 objectives on the Nova range and 21 of 31 on Lighthouse, close behind restricted GPT-6 Sol Daybreak, at under $3 per run versus about $22 — though no model completed a full attack chain. It matters because it shows inexpensive downloadable models can already carry out substantial parts of realistic intrusions, lowering the cost and barrier to autonomous cyberattacks.

• Alpha Signal2026-11-06 20:23
• Alpha Signal2026-10-20 09:08