Try GLM-5.2 in Claude Code Without Breaking Your Setup

If you're looking for a Fable 5 alternative to run inside Claude Code, this page shows how to try GLM-5.2 as a temporary replacement, with backup and rollback steps.

Right now, the highest-interest GLM-5.2 topics are long-context coding, Claude Code fallback setup after Fable 5, and quota-conscious routing: use GLM-5.2 for hard tasks, keep cheaper routing for lighter work.

If you are still evaluating GLM-5.2 before editing Claude Code settings, start with one of these external entry points. This page stays focused on the safe Claude Code setup path.

If you use Claude in the web app, desktop app, or mobile app, you probably do not need to edit any JSON files. Just switch to another available Claude model such as Opus or Sonnet.

This page is mainly for Claude Code CLI users who intentionally want to test GLM-5.2 through provider routing.

• Claude Code installed
• A provider account that supports GLM-5.2
• The correct API key
• The correct base URL or provider routing setup
• Access to edit ~/.claude/settings.json
• A backup of your current settings

The settings.json block below is not a complete provider setup. It only maps Claude Code aliases to GLM-5.2. Authentication and routing still depend on your provider setup.

cp ~/.claude/settings.json ~/.claude/settings.json.bak_$(date +%F)

If the file does not exist yet, create it first or follow the official Claude Code settings documentation for your environment.

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
    "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-4.5-air",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.2[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.2[1m]",
    "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "1000000"
  }
}

• This sets the required Z.ai base URL and API token fields.
• This maps Sonnet and Opus to glm-5.2[1m].
• This keeps Haiku on glm-4.5-air to avoid wasting quota.
• This is the safer default if you want 1M-context coding without routing every tiny task to the expensive model.
• Third-party provider behavior may differ from Anthropic-native models.
• Test this on a non-critical project first.

Because Claude Code uses smaller internal tasks too. If you route every alias to glm-5.2[1m], even lightweight helper work may burn higher-cost quota.

The default on this page is a quota-conscious setup: keep the stronger long-context model for Sonnet and Opus style work, and leave Haiku on a cheaper, faster route.

After editing settings, restart Claude Code and run /status to inspect the active model and provider details.

If the output does not match your expected provider or model, revert and check your API key, base URL, and provider documentation.

cp ~/.claude/settings.json.bak_YYYY-MM-DD ~/.claude/settings.json

Replace YYYY-MM-DD with your actual backup date.

• GLM-5.2 is not an official Anthropic replacement for Fable 5.
• This page is independent and source-backed.
• Provider behavior may differ in latency, context, and tool support.
• Model names, pricing, availability, and context behavior may change.
• Do not test this first on production-critical repositories.
• Always verify the latest provider details in official docs.

• Anthropic: Claude Code settings
• Anthropic: Claude Code model configuration
• Z.ai: How to switch models in Claude Code
• Z.ai: Claude Code overview

Last verified: June 15, 2026.

This page was written against the current Anthropic Claude Code settings docs and Z.ai Claude Code model-switch documentation available on that date.

This page is an independent setup note. It does not represent Anthropic, Z.ai, or any provider. Use it as a starting checklist, then confirm the latest behavior with official documentation before changing a real development environment.