Vals AI: DeepSeek V4.1 Flash 1017 beats GPT-6 Sol on offensive cybersecurity benchmark
Nunki08 · reddit · 2026-10-20
Vals AI's multi-step offensive cybersecurity eval shows DeepSeek V4.1 Flash 1017 averaging 16.3/21 objectives, edging out GPT-6 Sol Daybreak at 15.7/21 (tested via the Daybreak trusted access program). Grok 4.7 scored 12/21, Kimi K3 9/21, GLM 5.3 Flash 5/21, and Claude Opus 5.5 just 1/21 after refusing partway through. Measured by objectives completed per cumulative tokens (log scale), the top two models still burned well over 100M tokens on most runs.
Related event: DeepSeek V4.1 Flash Beats GPT-6 Sol in Vals AI Cybersecurity Benchmark(2 posts)→
More from Models
- Meta's Muse Confidential VM surfaces in web settings, subscriber-only at launch — testingcatalog · 2026-10-20
- DeepMind's four Gemini agents beat Minecraft in 27 hours, writing 2,783 programs and dying 605 times — ControlCAD · 2026-10-20
- DeepSeek-V4.1-Flash tops Agent Arena open models, ranks #11 overall at $0.20/task — vista8 · 2026-10-20
- DeepSeek V4.1 near Claude 5.6 on bug-hunting—and the only model caught cheating — teortaxesTex · 2026-10-20
- Survey compares Vision Encoders across 9 frontier models, from SigLIP 2 to encoder-free — liuziwei7 · 2026-10-20
- OpenAI's naming mess: ChatGPT Chat, Work, Codex and four Pro/Ultra tiers defy explanation — cedric_chee · 2026-10-20