Skip to content
CC Switch WikiUnofficial Community Wiki

Local qwen3-coder model connection fails with 502 error - cc-switch

Quick fix

Claude Code requires Anthropic Messages; use a format-conversion proxy such as LiteLLM between cc-switch and the local model.

Symptom

Section titled “Symptom”
Error output
CC-Switch发送的请求与这个成功的curl请求有差异，导致502错误。
English translation

Requests sent by CC-Switch differ from the successful curl request, causing a 502 error.

Cause

Section titled “Cause”

The root cause is incompatible API formats. Claude Code uses Anthropic's Messages API format, while the local qwen3-coder model deployed through llama-server provides an OpenAI-compatible Chat Completions API format. Although curl and Cherry Studio can call the OpenAI-format endpoint, Claude Code sends Anthropic-format requests, which the local server cannot parse and therefore returns a 502 error. The maintainer identified this as an output-format mismatch between chat format and anthropic messages format.

Fix

Section titled “Fix”

  1. Deploy a format-conversion proxy such as LiteLLM as an intermediary to convert Claude Code Anthropic-format requests to the OpenAI format supported by the local model. Example command: litellm --model openai/qwen3-coder --api_base http://localhost:8080/v1 --port 4000

    cc-switch 配置\n{
      "env": {
        "ANTHROPIC_BASE_URL": "http://localhost:4000",
        "ANTHROPIC_MODEL": "qwen3-coder"
      }
    }

  2. In cc-switch, point ANTHROPIC_BASE_URL to the local conversion proxy address, such as http://localhost:4000, instead of directly to the OpenAI-compatible endpoint of llama-server.

    cc-switch 配置\n{
      "env": {
        "ANTHROPIC_BASE_URL": "http://localhost:4000"
      }
    }

  3. Keep the local llama-server running. The proxy layer converts the format and forwards requests to http://localhost:8080/v1.

Affected Versions

Section titled “Affected Versions”
Toolclaude-code
VersionUnknown
PlatformsmacOSLinuxWindows

Source Issues

Section titled “Source Issues”

This page is distilled from 1 real issues

FAQ

Section titled “FAQ”
Why can curl and Cherry Studio call the local model successfully, but cc-switch cannot?
curl and Cherry Studio use the OpenAI-compatible Chat Completions format, while Claude Code configured through cc-switch sends the Anthropic Messages format. The formats are incompatible.
Can I configure the OpenAI-compatible endpoint of local llama-server directly in cc-switch?
No. Claude Code requires the Anthropic format. Configuring an OpenAI endpoint directly causes a format mismatch and a 502 error, so a format-conversion proxy is required.
Section titled “Related problems”
ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYRun `claude daemon stop --any` to stop lingering Claude background processes, then restart and resume the session.1 issues mergedBash failed due to the following issueSet the local GLM provider to OpenAI Chat format and enable routing so CC Switch converts and forwards Claude Code requests.2 issues mergedBoth ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY setSwitching from DeepSeek or MiMo to Codex writes both authentication variables at once; this issue has no confirmed Codex fix.1 issues merged

This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.