Mixed Model Collaboration

Four Agents powered by different models collaborate on a complex task, demonstrating the design philosophy of "different models for different roles."

Estimated time: 5 minutes
Number of Agents: 4 (project manager + analyst + developer + copywriter)
Models: Claude + DeepSeek + MiniMax

Design rationale

Different models excel in different areas. A mixed squad can optimize for both capability and cost:

Agent	Model	Why this model	Cost
项目经理	DeepSeek	Strong reasoning + orchestration	Low
分析师	Claude	Best at complex analysis	High
开发	DeepSeek	Strong coding ability	Low
文案	MiniMax	Good Chinese, extremely low cost	Very low

Steps

1. Create Agents

# Project manager — DeepSeek (orchestration)
ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic \
ANTHROPIC_AUTH_TOKEN=你的DeepSeek-Key \
anet node create 项目经理 --runtime claude-agent-sdk

# Analyst — Claude (deep analysis)
ANTHROPIC_API_KEY=sk-ant-xxx \
anet node create 分析师 --runtime claude-agent-sdk --model claude-sonnet-4-6

# Developer — DeepSeek (coding)
ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic \
ANTHROPIC_AUTH_TOKEN=你的DeepSeek-Key \
anet node create 开发 --runtime claude-agent-sdk

# Copywriter — MiniMax (documentation)
ANTHROPIC_BASE_URL=https://api.minimaxi.com/anthropic \
ANTHROPIC_AUTH_TOKEN=你的MiniMax-Key \
anet node create 文案 --runtime claude-agent-sdk

2. Start

anet node start 项目经理
anet node start 分析师
anet node start 开发
anet node start 文案

3. Send a task

anet task send 项目经理 "任务：设计并实现一个简单的待办事项 API。请分工：1）让分析师分析需求和技术方案；2）让开发写代码；3）让文案写 API 文档。最后汇总给我。网络中的 Agent：分析师、开发、文案。"

Architecture

              ┌──────────┐
              │ 项目经理  │ ← DeepSeek (orchestration)
              └─────┬────┘
                    │
       ┌────────────┼────────────┐
       ▼            ▼            ▼
 ┌──────────┐ ┌──────────┐ ┌──────────┐
 │  分析师   │ │   开发    │ │   文案    │
 │ (Claude)  │ │(DeepSeek)│ │(MiniMax) │
 │Deep analysis│ │  Coding  │ │  Docs    │
 └──────────┘ └──────────┘ └──────────┘
    High cost     Low cost    Very low cost

Cost comparison

Assuming an average of 10K tokens per task:

Approach	Approximate total cost
All Claude	~$0.12
Mixed squad	~$0.03
All MiniMax	~$0.005

The mixed squad uses a strong model for critical work (analysis) and cheaper models for routine work (development, copywriting) -- 75% cost reduction while maintaining quality.

Next steps

• Multi-Model Configuration -- Learn about all supported models
• Telegram Squad -- Large-scale Docker orchestration