DeepSeek V4.1 Flash 1017 tops open-source Cyber Index at 57%

Artificial Analysis published a Cyber Index update on October 24: the new 1017 version of DeepSeek's flagship open-weights model V4.1 Flash saw its cyber defense evaluation score jump from 41% to 57% (+16 points), making it the top-ranked openly available model on the index and second overall, behind only GPT-6, which is offered via a trusted access program. The evaluation also provided cost data.

Confirmed

• DeepSeek V4.1 Flash 1017 scored 57% on the Cyber Index, up 16 points from the previous version's 41%, taking the top spot among openly available models and surpassing rivals such as Grok (per Artificial Analysis)
• Second overall, behind only GPT-6 (offered via a trusted access program, not openly available)
• Cost per task is $0.42, more than double MiMo-V2.6-Pro (56 points, $0.18), but roughly 4x cheaper than GPT-6 Sol (Daylight Blue version)
• Cyber Index methodology: combines three evaluations based on real software repositories for vulnerability discovery, reproduction, and remediation — CWE-Bench-AA (based on Collinear AI's CWE-bench), DeepsecBench-A, and others

Why It Matters

• For the first time, an open-weights model is approaching the strongest closed-source models on cybersecurity defense tasks at roughly a quarter of the cost, offering strong value for security engineering automation
• The index is based on real-codebase tasks rather than synthetic problems, making the results a more reliable reference for practical vulnerability remediation capability

2026-10-24 ~ 2026-10-24 · 5 related posts

• DeepSeek V4.1 Flash 1017 tops open models on Artificial Analysis Cyber Index at 57% — ArtificialAnlys · 2026-10-24
• [source] DeepSeek V4.1 Flash 1017 Cyber Index results: $0.42/task, 4x cheaper than GPT-6 Sol — ArtificialAnlys · 2026-10-24
• [source] Inside the Artificial Analysis Cyber Index: three real-repo vuln-fixing evals, DeepSeek on top — ArtificialAnlys · 2026-10-24

2 near-duplicate retellings: ArtificialAnlys · ArtificialAnlys