Vals AI: DeepSeek V4.1 Flash Beats GPT-6 Sol on Offensive Cybersecurity Benchmark
Hesamation · x · 2026-10-20
Third-party evaluator Vals AI shared the first finding from its cybersecurity benchmark, built on simulated corporate networks: DeepSeek V4.1 Flash 1017 beats GPT-6 Sol (tested via the Daybreak trusted access program) on multi-step offensive cybersecurity tasks. A chart shows DeepSeek and GPT-6 Sol ahead of Grok 4.7, Kimi K3, GLM 5.3 Flash, and Claude Opus 5.5. The author stresses that offensive cyber capabilities OpenAI kept gated are now matched by a Chinese open model anyone can run locally. Full results will be published in the coming weeks; the benchmark aims to gauge the cyber risk a model poses if misused.
More from Models
- Cursor Pro+ $60/month quota shown: Cursor Models 25% used, Other Models 21% — pramodk73 · 2026-10-20
- "OpenAI really doesn't have the capacity": model hits capacity error mid-task — cedric_chee · 2026-10-20
- Same bug hits Astra 6, Sol 6.1 and Luna 6, but not Terra 5.6 — timigod · 2026-10-20
- DeepSeek V4.1 Flash uses 4.3x tokens and 2.4x time vs GPT-6 Sol, gap within noise — teortaxesTex · 2026-10-20
- Dev reports 5 hours of Codex outages while OpenAI status page stays all green — timigod · 2026-10-20
- Stratechery: Meta's new open model, the surprise Apple partnership, and Thin vs Thick — Stratechery · 2026-10-20