AI Bill of Materials
Model cards
Every model Securie runs in production. License, provider, chain assignment, and most-recent measured F1 — sourced from Securie's production inference configuration and a held-out public benchmark. Every measurement on this page is version-controlled and ships through code review: an updated number requires a reviewed change.
Foundation-Sec-8B-Reasoning
Cisco · localLicense:
Llama 3.1 Community License· Parameters: 8B · Chains: secret-scanMeasured Cyber benchmark composite = outperforms Llama 70B baseline on CyberMetric (Cisco internal) (2025-12)
GLM-5.1
Zhipu AI · oss-apiLicense:
MIT· Parameters: Mixed (not disclosed) · Chains: auth-authzdefaultMeasured auth-authz F1 = 1.000 on Securie public-bench F3 (208 fixtures) (2026-04-29)
Hermes 4 405B
NousResearch · oss-apiLicense:
MIT· Parameters: 405B · Chains: supabase-rlsbola-bflaidorMeasured bola-bfla F1 = 1.000 on Securie public-bench held-out (n=8 bola-bfla) (2026-04-29)
Gemini 2.5 Flash-Lite
Google · frontierLicense:
Proprietary (Google Cloud API)· Parameters: Mixed (not disclosed) · Chains: secret-scan-fallbackMeasured secret-scan F1 = 1.000 on canonical fixtures on Securie public-bench secret-scan (2026-04-29)
Claude Sonnet 4.6
Anthropic · frontierLicense:
Proprietary (Anthropic API)· Parameters: Mixed (not disclosed) · Chains: frontier-last-resortMeasured blended F1 = 0.952 on Securie public-bench held-out (cross-class) (2026-04-29)