How to use Claude Code with an uncensored, unlimited model

1. 1

   Get an API key

   Create a key under API keys in your Unquant workspace. Any plan works; GLM-5.3-Flash is the stronger coding model, with the full 262K context on every tier.

2. 2

   Point Claude Code at Unquant

   Environment variables

   ANTHROPIC_BASE_URLhttps://api.unquant.io

   ANTHROPIC_AUTH_TOKENYour Unquant API key

   ANTHROPIC_MODELglm-5.3-flash

   CLAUDE_CODE_MAX_CONTEXT_TOKENS262144

   export ANTHROPIC_BASE_URL=https://api.unquant.io
   export ANTHROPIC_AUTH_TOKEN=your-unquant-key
   export ANTHROPIC_MODEL=glm-5.3-flash
   export CLAUDE_CODE_MAX_CONTEXT_TOKENS=262144

   $env:ANTHROPIC_BASE_URL = "https://api.unquant.io"
   $env:ANTHROPIC_AUTH_TOKEN = "your-unquant-key"
   $env:ANTHROPIC_MODEL = "glm-5.3-flash"
   $env:CLAUDE_CODE_MAX_CONTEXT_TOKENS = "262144"

3. 3

   Start Claude Code

   Run claude in your project. If you were signed in to a Claude account before, run /logout once so Claude Code uses your Unquant key, and make sure ANTHROPIC_API_KEY is not also set to an Anthropic key.

Make it permanent

Put the same values in ~/.claude/settings.json so every session uses them:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.unquant.io",
    "ANTHROPIC_AUTH_TOKEN": "your-unquant-key",
    "ANTHROPIC_MODEL": "glm-5.3-flash",
    "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "262144"
  }
}

How it behaves

• ModelsAny Claude model name Claude Code asks for, including its small background model, is served by the model on your plan. Set ANTHROPIC_MODEL to choose explicitly.
• Model namesKeep ANTHROPIC_MODEL set to glm-5.3-flash (or qwen-3.8-flash-next). With a Claude model name such as claude-opus-5-5, or any id ending in [1m], Claude Code assumes that model's 1M window, ignores CLAUDE_CODE_MAX_CONTEXT_TOKENS and stops compacting, so the session fails once it passes 262K tokens. The same applies behind a local proxy: give it the real 262,144-token window and never pick its 1M variant.
• ThinkingClaude Code's thinking setting maps to reasoning effort: off is none, small budgets low, larger ones high or max.
• ToolsFile edits, shell commands and every other Claude Code tool run through standard tool calling, with images passed through as well.
• ContextGLM-5.3-Flash serves 262,144 tokens on every plan, so whole-repository sessions fit before Claude Code needs to compact. With CLAUDE_CODE_MAX_CONTEXT_TOKENS set, Claude Code only adds to your history, so each turn reuses the cached conversation and starts fast.

Troubleshooting