Connecting Claude Code to Chinese AI Models: DeepSeek, Qwen, and GLM Configuration Guide

The official Anthropic API can be frustratingly unreliable in China — a pain point many developers know all too well. Fortunately, Claude Code's underlying protocol is open: by routing requests to a domestic platform that implements an Anthropic-compatible interface, you can seamlessly use models like DeepSeek, Alibaba's Qwen, and Zhipu's GLM. This guide walks through the setup for all three platforms and introduces CC Switch, a tool that lets you switch between configurations with a single click.

The Core Logic: Three Environment Variables Are All You Need

Regardless of which platform you're connecting to, the configuration pattern for Claude Code is consistent. Three environment variables do the heavy lifting:

• ANTHROPIC_BASE_URL: Specifies the API gateway address, redirecting requests from Anthropic's servers to your chosen platform.
• ANTHROPIC_AUTH_TOKEN: Your API key, used for authentication.
• ANTHROPIC_MODEL: Sets the default primary model.

Beyond this core trio, you can also fine-tune model mappings — assigning different models to the Opus, Sonnet, and Haiku tiers respectively. Use a lighter model for simple agent tasks, and set CLAUDE_CODE_EFFORT_LEVEL to Max to ensure the model runs at full capacity.

The prep work is straightforward: go to your chosen platform's console, register and complete identity verification, create an API key, and you're ready to start. There are two main ways to apply configuration — set environment variables directly, or write them into a settings.json file located in the .claude directory under your home folder.

Anthropic-compatible interfaces are proxy layers implemented by third-party platforms that mirror the request/response format of Anthropic's official API. As long as ANTHROPIC_BASE_URL points to an endpoint that supports this format, Claude Code can transparently call entirely different underlying models without any modification. This is similar to how OpenAI-compatible interfaces have become widespread in the industry — if the response structure matches, the client can't tell the difference on the backend. The reason all three Chinese platforms can integrate with Claude Code is precisely because they've each implemented this compatibility layer, wrapping models like DeepSeek^Ad-V4, Qwen3, and GLM into a format that Claude clients can recognize.

CLAUDE_CODE_EFFORT_LEVEL is a reasoning intensity parameter specific to Claude Code. When set to Max, the model engages deeper chain-of-thought processing, which is ideal for complex code generation and debugging tasks — at the cost of longer response times and higher token consumption. When connecting to capable domestic models, this parameter can significantly improve output quality.

DeepSeek: Set Environment Variables via PowerShell

Of the three platforms, DeepSeek is best configured through environment variables. In PowerShell, set ANTHROPIC_BASE_URL to https://api.deepseek.com/anthropic, replace ANTHROPIC_AUTH_TOKEN with your DeepSeek API key, and set ANTHROPIC_MODEL to DeepSeek-V4-Pro.

Two common pitfalls to watch out for: the model name needs the [1M] suffix to indicate context length, and remember to set the reasoning effort to Max to unlock the model's full capability.

Once configured, navigate to your project directory and run claude to start. You can also specify the model at launch with claude --model deepseek-v4-pro, or switch models within an interactive session by typing /model. Typing /model also displays the list of currently supported models.

Qwen and GLM: settings.json Is the Cleaner Option

For Alibaba's Qwen via the Alibaba Cloud Bailian platform, the configuration file approach is recommended. Open the .claude folder in your home directory, edit settings.json, and add your ANTHROPIC_AUTH_TOKEN (your API key), ANTHROPIC_BASE_URL (https://dashscope.aliyuncs.com/apps/anthropic), and ANTHROPIC_MODEL (qwen3-coder-plus) under the env field. Save the file and reopen your terminal for the changes to take effect.

Windows users can alternatively use set in CMD or write user-level environment variables in PowerShell — the result is the same. To switch models mid-session, type /model qwen3-coder-plus in the conversation.

Zhipu GLM follows the same settings.json pattern under the env field: set ANTHROPIC_AUTH_TOKEN to your API key and ANTHROPIC_BASE_URL to https://open.bigmodel.cn/api/anthropic. You can also extend the API timeout (e.g., to 3,000,000 milliseconds) and set an environment variable to disable unnecessary traffic, reducing unwanted data transmission. Once configured, run claude, then type /status to confirm the active model. If it's not correct, use /config to switch.

CC Switch: One-Click Multi-Platform Switching

When you're juggling multiple platforms, constantly editing configuration files becomes a chore. CC Switch is a third-party tool worth recommending — it supports Windows, macOS, and Linux, and serves as a companion manager for Claude Code, Codex, and Gemini CLI. Think of it as a smart organizer for your development toolkit.

Installation is simple: search for firon1231/ccswitch on GitHub (verify the exact repo name), then download the installer for your platform from the Releases page. Once installed, switching between API configurations across different platforms is just a click away — no more manual environment variable editing. For more detailed usage instructions, refer to dedicated CC Switch articles available online.

From a technical standpoint, CC Switch is essentially a configuration file manager. It stores multiple sets of environment variable combinations (BASE_URL + AUTH_TOKEN + MODEL) as named "profiles" and uses atomic operations to swap the active variable set when you switch, syncing changes to your shell config or system environment variables so Claude Code automatically loads the target platform's settings on next launch. For teams that use different models across different projects, this approach eliminates the typos and missing fields that come with manual config file editing.

Quick Reference and Configuration Patterns

Key details for all three platforms at a glance:

Platform	Gateway URL	Default Model	Recommended Method
DeepSeek	api.deepseek.com/anthropic	DeepSeek-V4-Pro [1M]	Environment variables
Alibaba Bailian	dashscope.aliyuncs.com/.../anthropic	qwen3-coder-plus	settings.json
Zhipu	open.bigmodel.cn/api/anthropic	GLM series	settings.json

The whole process comes down to one simple rule: get your API key from the platform console, configure the URL + Token + Model trio, launch Claude, and verify with /status or /model. If you're managing multiple platforms, CC Switch handles it all in one click.

The next installment will take a deeper dive into DeepSeek's configuration options and explore more advanced usage patterns.