Qwen Code

Qwen Code reads its endpoints from modelProviders in ~/.qwen/settings.json — one entry with a baseUrl and an envKey puts Claude, GPT and Gemini in its /model picker.

{
  "modelProviders": {
    "openai": [
      {
        "id": "claude-sonnet-5",
        "name": "Claude Sonnet 5 (Kunavo)",
        "baseUrl": "https://api.kunavo.com/v1",
        "description": "Kunavo, OpenAI-compatible",
        "envKey": "KUNAVO_API_KEY"
      }
    ]
  },
  "env": {
    "KUNAVO_API_KEY": "sk-kn-..."
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "model": {
    "name": "claude-sonnet-5"
  }
}

Step by step

1. Create a key at /app/keys and copy it — it is shown once.
2. Open ~/.qwen/settings.json (create it if it does not exist) and merge in the four blocks above. The docs recommend declaring modelProviders in the user-scope file “to avoid merge conflicts between project and user settings”.
3. Put the key somewhere better than env if you can. Qwen Code reads it from process.env[envKey], and the docs rank the sources highest to lowest: a shell export, then a .env file, then the env block in settings.json — which they flag as plain-text storage. The env block above is the smallest thing that runs, not the best thing to keep.
4. Run qwen. With security.auth.selectedType set to openai and model.name matching an id you declared, no interactive /auth step is needed — the docs say so in as many words after the one-file example.
5. Give it a task that reads and edits a file, not a greeting. Qwen Code is an agent: tool calling and streaming are what a first run should exercise, and they are what would fail first against an endpoint that only half fits.
6. Add more entries under modelProviders.openai to switch models at runtime with /model. Those edits hot-reload in a running session; providerProtocol is read once at startup and needs a restart.

Checked against Qwen Code's authentication page, Option 4: API Key (flexible) on September 21, 2026. Third-party settings move; if a field name here no longer matches what you see, that page is the authority, not this one.

Verify before you debug the client

One request settles whether a failure is the endpoint, the key, or the configuration file. If this returns JSON, the same base URL and key work in Qwen Code.

# Settles whether a failure is the endpoint, the key, or the client.
curl -sS https://api.kunavo.com/v1/models \
  -H "Authorization: Bearer sk-kn-..."

Which model id to put in the field

Every text model is reachable as a model id — the live list is GET /v1/models, and the catalog with prices is on the models page. Rates are USD per 1M tokens, input / output.

Three things the docs settle that people guess wrong

These come from the authentication page and the model providers reference linked above, and each one costs real debugging time when it is assumed instead of read.

1. A modelProviders entry outranks the CLI flags. The documented order, highest first, is: overrides made through /auth in the running session, then the selected model provider's envKey, then CLI arguments such as --openai-api-key, then environment variables, then security.auth.apiKey in settings. Most people expect the flag to win. It does not — which is why --openai-base-url can appear to be ignored.
2. security.auth.apiKey and security.auth.baseUrl are deprecated. The reference says so and recommends migrating to modelProviders. If an older tutorial has you editing those two keys, it is editing the route that is on its way out.
3. wireApi picks the request format, and nothing detects a mismatch. Omitting it means Chat Completions, which is what the block above uses. Setting "wireApi": "responses" requires a genuinely Responses-compatible endpoint, and the docs state plainly that no endpoint detection and no automatic fallback occur when a request fails. Kunavo answers /v1/responses as well as /v1/chat/completions, but this page has tested neither pairing, so start on the default.

If you came here looking for the free tier

Much of what is still written about Qwen Code describes a Qwen OAuth login with a free daily allowance. That option is gone: the documentation records its free tier as discontinued on April 15, 2026, and says Qwen OAuth is no longer a selectable entry in the /auth dialog. The three it lists now are Alibaba ModelStudio — with Coding Plan, Token Plan and Standard API Key in its sub-menu — Third-party Providers, and Custom Provider, described as connecting “a local server, proxy, or unsupported provider”. Kunavo is the third of those. Note also that the ModelStudio sub-menu items are not three ways to pay one bill: each has its own host and its own key, and a Coding Plan key on a Token Plan host will not work.