Tools Model Security Capability claude-sonnet-4-6
Model record
Claude Sonnet 4.6: how it does on security work, and what it will answer
- Lab
- Anthropic
- Access
- api
- Weights
- closed
- Safeguards
- Real-time cyber safeguards on Sonnet-class models
- Score rows
- 1
- Confidence
- CONFIRMED
What happened
Carried for one row: Kolega's own Sonnet 4.6 adaptation on RealVuln scored F3 47.9 on the Python subset at 66.89% precision, and at $0.2173 per 100 lines it is the most expensive row on the board per line.
RealVuln 3.1.0, F3 (micro): 47.9 (independent; Kolega adaptation on Claude Code (kolega-ca-cc-sonnet); 66 of 140 repositories pinned by commit SHA (Python subset); safeguards: production safeguards as deployed on the API; run 2026-09-11; CONFIRMED). strict F3 22.9; precision 66.89%, recall 46.42%; $290.68 for the run; $0.2173 per 100 lines; 396.6s wall clock a run. Refusal gate (CONFIRMED, observed 2026-09-11): Real-time cyber safeguards on Opus and Sonnet.
Sources
- RealVuln dashboardraw.githubusercontent.com/kolega-ai/Real-Vuln-Benchmark/main…
- Refusal policy or licenceplatform.claude.com/docs/en/build-with-claude/refusals-and-f…
On YFarmX
- Reference pageyfarmx.com/ai/llms/claude-sonnet-4-6/
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