| Type |
Gateway + native cyber LLM |
Hosted abliterated LLM |
Specialist cyber lab (NJ LLC) |
| Offensive generation |
✅ gated Zero + uncensored Qwen |
✅ no per-request refusals |
✅ Kimi-K3-ablated + cyber-tuned, auth-first ToS |
| Models |
200+ routable, 1 native cyber + 3+ uncensored |
3 hosted (base/large/large-v2) |
2 (lordx64/cyberkimi, cyberglm) + CyberSeek soon |
| Ctx |
Zero 1M/128K out; Qwen 262K |
Base 262K; Large 1M/1M |
Kimi 1M total (750K client, 512K bench); GLM 131K client |
| Cyber bench (vendor) |
Zero 98.07% CyberGym L1 |
Large-v2 84.5% CyberGym, 2× ExploitBench |
Kimi 86.7% CyberGym (78/90); 10/16 assisted V8 CVE-2024-6100 + public transcripts |
| Entry price |
$0 free + $3/$5 Zero |
$20/mo + $3/$3, $5/$5 |
PAYG wallet only (no current sub); top-up $10+ |
| Scale cost |
~$36/mo per 10M (70% in) |
~$50-80/mo per 10M + sub |
Kimi $8/$0.80/$30; GLM $4/$0.40/$15 — ~$200+/10M output-heavy |
| API |
api.orcarouter.ai/v1 OpenAI + Anthropic/Gemini, MCP, Lite |
api.abliteration.ai/v1 OpenAI/Anthropic, Promptfoo/Garak/Mastra |
api.adverserial.ai/v1 OpenAI + Anthropic shim, Claude/Codex/OpenCode/Hermes docs |
| Logging / audit |
Receipts/logs, budgets/roles, Team compliance |
Zero-retention default, Gateway Ent-only |
No inference/chat logs claim; billing + KV-cache retained; DPA/region only via Enterprise |
| Guardrails |
Engagement+passkey+terms gating, agent firewall |
Customer Gateway (“Unrestricted. Not ungoverned”) |
Auth-first ToS, human-review mandatory, “capability ≠ permission” |
| Jurisdiction |
Singapore SIAC |
Delaware |
New Jersey LLC |
| Best for |
Governed multi-model routing + cheap triage |
Cheap uncensored volume |
Hard repro + evidence-backed reports |
