Three-Layer Framework LLM API Configuration Summary: DS V4 Pro, Qwen 3.6, oMLX Local

Model Configuration Overview by Layer

Hermes Agent         OpenClaw                Claude Code
──────────────────────────────────────────────────────────────────
Model        Qwen 3.6-Plus        DS V4 Pro ✅             DS V4 Pro ✅
           (ModelStudio)         Qwen 3.6-Plus ✅         Qwen 3.6-Plus ✅
oMLX Qwen3 Coder ✅      Local Qwen3 Coder ✅
──────────────────────────────────────────────────────────────────
Purpose        Background collection              Primary conversation                 Multi-instance code
           System checks              Task orchestration                 · Pro: complex logic
           Skill learning              Skill scheduling                 · Qwen: rapid iteration
           (Backup/standby)            Deployment decisions                 · Local: private code

---

Hermes Agent: Qwen 3.6-Plus

# ~/.hermes/config.yaml
model:
  provider: modelstudio
  model: qwen3.6-plus
  api_key: "sk-xxx"
  base_url: "https://dashscope.aliyuncs.com/compatible-mode/v1"
  plan: coding  # ModelStudio Coding Plan

---

OpenClaw: Three Backends

// ~/.openclaw/openclaw.json
{
  "models": {
    "default": "deepseek-pro",
    "providers": {
      "deepseek-pro": {
        "api_key": "sk-xxx",
        "base_url": "https://api.deepseek.com/v1",
        "model": "deepseek-reasoner"
      },
      "qwen": {
        "api_key": "sk-xxx",
        "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
        "model": "qwen3.6-plus"
      },
      "local-coder": {
        "api_key": "not-needed",
        "base_url": "http://localhost:8080/v1",
        "model": "qwen3-coder-next"
      }
    },
    "routing": {
      "rules": [
        {"pattern": "code|refactor|debug", "provider": "deepseek-pro"},
{"pattern": "Chinese|translate|content", "provider": "qwen"},
        {"pattern": "private|secret|key|password", "provider": "local-coder"}
      ]
    }
  }
}

---

oMLX Local Qwen3 Coder Next

Launch

# Download Model (MLX Format)
mlx_lm.convert --hf-path Qwen/Qwen3-Coder-Next --mlx-path ~/models/qwen3-coder-next -q

# Start API Server
mlx_lm.server --model ~/models/qwen3-coder-next --port 8080

Configure the Claude Code Local Instance

// ~/.claude-local.json
{
  "apiKey": "not-needed",
  "baseURL": "http://localhost:8080/v1",
  "model": "qwen3-coder-next"
}

Launch:

claude --config ~/.claude-local.json --cwd ~/my-private-project

---

Managing the Three Claude Code Instances

#!/bin/bash
# ~/bin/cc-launcher

case "$1" in
  pro)
    export CLAUDE_CONFIG=~/.claude-pro.json  # DS V4 Pro
    claude --cwd "$2"
    ;;
  qwen)
    export CLAUDE_CONFIG=~/.claude-qwen.json  # Qwen 3.6-Plus
    claude --cwd "$2"
;;
  local)
export CLAUDE_CONFIG=~/.claude-local.json # oMLX local
    claude --cwd "$2"
    ;;
  *)
    echo "Usage: cc-launcher [pro|qwen|local] [project-dir]"
    ;;
esac

alias cc-pro='cc-launcher pro'
alias cc-qwen='cc-launcher qwen'
alias cc-local='cc-launcher local'

---

Cost Comparison

Model	Input / 1M tokens	Output / 1M tokens	Strengths
DS V4 Pro	$0.55	$2.19	Reasoning, architecture
DS V4 Flash	$0.14	$0.28	Everyday use, speed
Qwen 3.6-Plus	~$1.00	~$4.00	Chinese language, content
Local Qwen3 Coder	Free	Free	Private code

---

Related Articles

• Agent Framework Capability Matrix
• oMLX + Qwen3 Coder Next Deployment
• DeepSeek V4 Guide