Skip to content

model-toggle.md

Latest commit

 

History

History
90 lines (74 loc) · 3.17 KB

model-toggle.md

File metadata and controls

90 lines (74 loc) · 3.17 KB
Raw

Model Toggle

The pack defaults to multi-model mode — per-agent model orchestration where each agent uses the opencode-go model best suited to its task. You can switch to single-model mode where all agents inherit one model.

Why two modes?

  • Multi-model (default): best capability/cost balance. glm-5.2 for build, deepseek-v4-pro for planning, qwen3.7-max for review, deepseek-v4-flash for cheap exploration. Maximizes quality while controlling cost.
  • Single-model: simpler to reason about. Everything runs on one model. Useful for benchmarking, debugging, or when you want simplicity.

Commands

Windows (PowerShell)

.\toggle-models.ps1 status                              # show current mode + agent models
.\toggle-models.ps1 multi                               # restore per-agent model assignments
.\toggle-models.ps1 single                              # all agents use top-level model
.\toggle-models.ps1 single opencode-go/kimi-k2.7-code   # single + pick specific model

macOS / Linux

./toggle-models.sh status
./toggle-models.sh multi
./toggle-models.sh single
./toggle-models.sh single opencode-go/kimi-k2.7-code

How it works

Multi mode

The script reads profiles/multi-models.json and writes per-agent model fields into opencode.json:

{
  "model": "opencode-go/glm-5.2",
  "small_model": "opencode-go/deepseek-v4-flash",
  "agent": {
    "build":             { "model": "opencode-go/glm-5.2" },
    "plan":              { "model": "opencode-go/deepseek-v4-pro" },
    "reviewer":          { "model": "opencode-go/qwen3.7-max" },
    ...
  }
}

Single mode

The script removes all per-agent model fields from opencode.json. Every agent inherits the top-level model:

{
  "model": "opencode-go/glm-5.2",
  "small_model": "opencode-go/glm-5.2",
  "agent": {
    "build":             {},
    "plan":              {},
    "reviewer":          {},
    ...
  }
}

Customizing multi-mode assignments

Edit ~/.config/opencode/profiles/multi-models.json to change which model each agent uses in multi mode. The toggle script reads this file on every multi invocation.

{
  "model": "opencode-go/glm-5.2",
  "small_model": "opencode-go/deepseek-v4-flash",
  "agents": {
    "build": "opencode-go/glm-5.2",
    "plan": "opencode-go/deepseek-v4-pro",
    "reviewer": "opencode-go/qwen3.7-max",
    "explorer": "opencode-go/deepseek-v4-flash",
    "data-engineer": "opencode-go/kimi-k2.7-code",
    "frontend-engineer": "opencode-go/glm-5.2",
    "devops-engineer": "opencode-go/deepseek-v4-pro",
    "doc-writer": "opencode-go/mimo-v2.5",
    "title": "opencode-go/deepseek-v4-flash",
    "summary": "opencode-go/deepseek-v4-flash",
    "compaction": "opencode-go/deepseek-v4-flash"
  }
}

Ad-hoc switching (without the script)

opencode's TUI /model command switches the active session's model on the fly — useful for one-off switches without editing config. The toggle script is for persistent config-level switching across all agents and future sessions.

After toggling

Restart opencode for config changes to take effect. Config is loaded once at startup and is not hot-reloaded.