Model Selection

How to choose models for different agent roles.

Model Tiers

Tier	Example Models	Token Cost	Best For
Premium	deepseek-v4-pro, claude-sonnet-4	$$	Architecture, debugging, complex code
Medium	glm-5.1, minimax-m2.7	$	Exploration, research, moderate tasks
Free	minimax-m2.5-free, big-pickle	Free	Quick fixes, formatting, simple changes

Fallback Chains

If a model is unavailable or rate-limited, the agent falls through the chain:

{
  "model": "opencode-go/deepseek-v4-pro",
  "fallback_models": [
    { "model": "opencode-go/glm-5.1" },
    { "model": "opencode-go/minimax-m2.7" },
    { "model": "opencode-go/minimax-m2.5-free" }
  ]
}

Cost-Optimized Setup

Route more work to free/cheap models:

• quick → minimax-m2.5-free
• writing → minimax-m2.5-free
• explore, librarian → glm-5.1
• sisyphus-junior → minimax-m2.7

This keeps premium models reserved for where they add the most value.

Quality-Optimized Setup

Use premium models across the board for the best output quality. Higher token cost but fewer iterations and better results.