Claude Code
Models, options, and features for Claude Code.
Send a claude_code block to run a session with Claude Code:
JSON
1
{
2
"claude_code": {
3
"prompt": "Find and fix the failing test.",
4
"model": "claude-opus-5-5",
5
"effort": "high"
6
}
7
}
In agent YAML, the same fields go under session.claude_code. The CLI picks the harness from --model.
Options
The task.
A Claude Code model from Models.
Reasoning effort, where the model supports it.
The most turns the agent can take on each message, a whole number above 0.
The model Claude Code switches to when model is overloaded. Default none.
A Claude Code settings.json in a repository, applied below the settings Ellipsis manages. Hooks stay disabled.
| Model | Effort levels | Default effort |
|---|---|---|
| Claude Opus 5.5 and Sonnet 5.5 | All 5 | medium |
| Claude Opus 4.7 | All 5 | xhigh |
| Claude Fable 5.1, Fable 5, Opus 5, and Opus 4.8 | All 5 | high |
| Claude Opus 4.6 and Sonnet 4.6 | All except xhigh | high |
| Claude Haiku 4.5, GLM-5.2, and MiniMax M3 | None | None |
Features
| Feature | Support |
|---|---|
| Follow-up messages | All models |
| MCP servers | All models |
| JSON output | Claude models; not GLM-5.2 or MiniMax M3 |
| Skills | Not supported; use Codex |
| Images in messages | Not supported; use Codex |
Subscription
To pay with your Claude plan instead of credits, connect your Claude subscription. A session keeps the token it started with; a replaced token applies to new sessions.