CUTEADMOA-5.1 Documentation

Production-ready Mixture of Agents with 5 NVIDIA models. 6 dedicated API keys with per-model distribution.

Architecture

Two-layer MOA architecture:

1. Layer 1 — Parallel Execution: All 5 models fire simultaneously, each on its own dedicated API key
2. Layer 2 — Aggregation: Nemotron 550B synthesizes all responses into one superior answer

Models (5 — Per-Model Keys)

Key	Model	Provider	Max Tokens	Role
nemotron	Nemotron 3 Ultra 550B	NVIDIA	16,384	Reasoning / Aggregator
minimax	Minimax M3	Minimax	8,192	General / Fast
glm52	GLM 5.2	Z.ai	16,384	Bilingual
deepseek	DeepSeek V4 Pro	DeepSeek AI	16,384	Deep Reasoning
step37	Step 3.7 Flash	StepFun	16,384	Fast Reasoning

API Reference

POST /CUTEADMOA5.1/api/generate

Generate a response using the MOA pipeline. Each model uses its own API key.

Parameter	Type	Required	Description
task	string	Yes	The task or prompt
mode	string	No	"parallel" (default) or "sequential"
models	array	No	nemotron, minimax, glm52, deepseek, step37. Default: all 5

Example

curl -X POST https://cuteadmoa.site/CUTEADMOA5.1/api/generate \
  -H "Content-Type: application/json" \
  -d '{"task": "Build a REST API in FastAPI","mode":"parallel","models":["nemotron","minimax","glm52","deepseek","step37"]}'

Response

{
  "success": true,
  "result": {
    "aggregated": "Full synthesized response...",
    "summary": "Key insights in bullet points...",
    "individual": [
      {"model":"nemotron","model_name":"Nemotron 3 Ultra 550B","content":"...","success":true},
      ...
    ]
  },
  "stats": {"total":5,"successful":5,"failed":0,"mode":"parallel","keys_used":6},
  "timestamp": "2026-07-29T..."
}

GET /CUTEADMOA5.1/api/health

curl https://cuteadmoa.site/CUTEADMOA5.1/api/health
# {"status":"ok","brand":"CUTEADMOA-5.1","version":"5.1.0","keys":6,"key_mode":"per-model","models":["nemotron","minimax","glm52","deepseek","step37"]}

Modes

Parallel

All models run simultaneously. ~time of slowest model.

Sequential

Models run one after another. Later models see previous responses.

6-Key Distribution

• 5 models → 5 dedicated keys (no shared pressure)
• 1 aggregator key → with fallback across all 6 keys
• Rate limit isolated to individual model keys

Python Integration

import requests
r = requests.post("https://cuteadmoa.site/CUTEADMOA5.1/api/generate",
    json={"task": "Build a GraphQL API", "mode": "parallel"})
data = r.json()
print(data["result"]["summary"])       # Key insights
print(data["result"]["aggregated"])    # Full response
for m in data["result"]["individual"]:
    print(f"{m['model_name']}: {m['content'][:80]}...")