🚀 MAD HARE v1.0.0 GA LAUNCH Built solo by 1 founder with an autonomous agent fleet. Live tonight!
● Cloud Run + AlloyDB Live • Client #1: JBC Active
EVAL.MADHARE.AI TENANT BENCHMARK GATEWAY

Model Evaluation & Pareto Frontier

Multi-tier agent benchmark scoring for tenant JBC (JBC3 Platform)

🛠️ Support & Docs ← Back to Agent Portal
Model Pareto Frontier (Cost vs Latency vs Quality)
Model & Infrastructure Role in Conflux DAG Cost / 1K Tok P95 Latency Quality Score
judge-qwen3-8b-awq (Local port 18901) Tier 2 SLM Triage Gatekeeper $0.0000 38 ms 94.2%
gemini-2.5-flash (Google Vertex AI) Tier 1 Central Planner & Dispatcher $0.00015 180 ms 97.8%
claude-3-5-sonnet (Max CLI OAuth) Tier 3 Sandboxed Code Reviewer $0.0030 820 ms 99.1%
AlloyDB Omni & ScaNN SQ8 Vector Search Benchmarks
Blueprints Indexed
151
Vector Cosine Recall
99.4%
Query Latency
1.2 ms
Quantizer
SQ8