benchmarks.json

{
  "model_name": "EAM-100M-Agentic-Kernel",
  "version": "1.1 (TurboQuant + IndexCache)",
  "category": "Edge Agentic Model (Sub-200M)",
  "benchmark_date": "2026-05-04",
  "benchmarks": {
    "ARC-Challenge (8-Shot Reasoning)": {
      "accuracy": "20.0%",
      "status": "Verified (Deterministic)",
      "throughput": "25.45 items/sec",
      "context_utilization": "1024 tokens"
    },
    "Agentic IQ v1.1": {
      "score": "66.7%",
      "discovery_awareness": "0/1",
      "fama_recovery": "1/1",
      "goal_persistence": "1/1",
      "avg_latency": "0.0073s",
      "improvement": "75% faster than v1.0"
    }
  },
  "optimizations": [
    "TurboQuant (Ternary Packing)",
    "IndexCache (Sparse KV-Caching)",
    "Greedy Search (Deterministic Reasoning)"
  ],
  "hardware_verified": "Local CPU (AMD/Intel)",
  "quantization_potential": "BitNet 1.58b (~30MB)"
}