Tools Model Security Capability deepseek-v4-pro
Model record
DeepSeek V4 Pro: how it does on security work, and what it will answer
- Lab
- DeepSeek
- Released
- 2026-08-13
- Access
- open-weights
- Weights
- open
- Safeguards
- DeepSeek licence lists twelve use restrictions; cyber activity is absent from the list
- Score rows
- 2
- Confidence
- CONFIRMED
What happened
RealVuln scores it F3 38.0 across all 140 repositories at 60.91% precision over three runs; a search-derived CVE campaign puts it at 87.5% pass@3 at 65.6% precision on 32 recent CVEs.
RealVuln 3.1.0, F3 (micro): 38 (independent; agentic harness (deepseek-v4-pro-agentic-v1), prompt sha256:45a1200d61e6; 140 of 140 repositories pinned by commit SHA (full corpus, TypeScript and JavaScript included); safeguards: open weights, no classifier; run 2026-09-11; CONFIRMED). strict F3 38.0; precision 60.91%, recall 36.43%; $15.08 for the run; $0.002 per 100 lines; 3 runs; 285.4s wall clock a run. Recent-CVE reproduction campaign, pass@3: 87.5% (independent; Dourassov, three runs; 32 recent CVEs; safeguards: open weights; run 2026-09; SINGLE). 65.6% precision; search-derived. Refusal gate (CONFIRMED, observed 2026-09-11): Licence use restrictions name nothing on intrusion or exploitation.
Sources
- RealVuln dashboardraw.githubusercontent.com/kolega-ai/Real-Vuln-Benchmark/main…
- Recent-CVE reproduction campaign sourceyfarmx.com/ai/llms/deepseek-v4-pro/
- Refusal policy or licenceraw.githubusercontent.com/deepseek-ai/DeepSeek-V3/main/LICEN…
On YFarmX
- Reference pageyfarmx.com/ai/llms/deepseek-v4-pro/
One record from the Model Security Capability, maintained by the Security Desk. Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 19 September 2026