typesafe/jev-1.13
LIVE
48,219 viewers
🛡️ Jev AutoMod Active
itisshikhar streamer camera
Team:
JEV COST
$0.000000
QUICK COMMS
[ BENCHMARK SPECIFICATION • INDIC TOXICITY SCREENING ]

Traditional Regex vs. TypeSafe AI Jev vs. OpenAI gpt-4o-mini

Evaluating production moderation architectures for high-volume Indic live gaming & stream chat (romanized Hinglish, Tanglish, Benglish, Kanglish, Tenglish). Comparing precision against evasion (leetspeak, symbol swaps, whitespace padding), inference latency, and token economics across scale.

Architecture Dimension 1. Regex / Keyword Filter 2. TypeSafe AI Jev (1.13) 3. OpenAI (gpt-4o-mini)
Core Mechanism Substring denylist / regex trie System One direct decision head Autoregressive generative LLM
OpenRouter Token Rate $0.00 (in-process) $0.042 in / $0 out per 1M $0.150 in / $0.600 out per 1M
Cost / 1,000,000 Checks $0.00 ~$4.20 (~100 tokens, $0 out) ~$55.00 - $185.00 (13x - 44x higher)
Cost / Single Decision $0.000000 ~$0.000004 ~$0.000055 - $0.000185
[ REGEX DENYLIST ] if (text.includes(kw))
--%
[OK] 0 correct [FAIL] 0 missed / wrong Cost: $0.00 • In-process trie Speed: <1ms (Instant)
[ TYPESAFE JEV 1.13 ] System One Decisions
--%
[OK] 0 correct [REVIEW] 0 routed [FAIL] 0 wrong Cost: $0.000000 Speed: ~350-500ms (Fast)
[ OPENAI GPT-4O-MINI ] Chat Completions JSON
--%
[OK] 0 correct [REVIEW] 0 routed [FAIL] 0 wrong Cost: $0.000000 Speed: ~1,500-2,500ms (Slower)