Claude Code on an uncensored model
Claude Code can talk to any server that speaks the Anthropic format. This site does, so it takes four environment variables.
The four variables
# macOS and Linux
export ANTHROPIC_BASE_URL=https://uncensoredchats.com
export ANTHROPIC_AUTH_TOKEN="sk-uc-..."
export ANTHROPIC_MODEL=glm-5.3-flash-xploded
export ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.3-flash-xploded
claude
# Windows PowerShell
$env:ANTHROPIC_BASE_URL = "https://uncensoredchats.com"
$env:ANTHROPIC_AUTH_TOKEN = "sk-uc-..."
$env:ANTHROPIC_MODEL = "glm-5.3-flash-xploded"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "glm-5.3-flash-xploded"
claudeThe base URL is the site itself with no /v1 on the end. Claude Code adds the path.
The setting most guides miss
Claude Code sends its small background jobs, such as naming the session, to a separate Haiku model. If you set only ANTHROPIC_MODEL, those calls still ask for a Claude model, which is not sold here, and fail. Setting ANTHROPIC_DEFAULT_HAIKU_MODEL to the same id fixes it.
What to expect
This is not a supported setup: Claude Code is built and tested for Claude models. Reading, editing and running commands work. Features tuned for Claude may behave differently, and long sessions can drift. For a tool designed to take any model, opencode is the smoother choice.
Which model
- GLM 5.3 Flash (
glm-5.3-flash-xploded) handles tool calls well and is cheap enough to leave running. - Kimi K3.1 Pro (
kimi-k3.1-pro-xploded) is the strongest, at $37.50 per million output tokens.
Before a long session
Put a spend cap on the key in the dashboard. Claude Code makes many requests per task, and a cap turns a runaway session into a stopped key.
Keep reading
- Kimi K3.1 Pro, uncensored→
- GLM 5.3, uncensored→
- DeepSeek V4.1 Flash, uncensored→
- Qwen3.8 27B, uncensored→
- The uncensored AI API→
- Set up SillyTavern→
- Set up opencode→
- Abliterated vs uncensored→
- Cost calculator→
- Uncensored AI roleplay→
Updated 12 November 2026. Prices are read from the live price list.