Qwen Code

Connect Qwen Code — the terminal agent from the Qwen team (a Gemini CLI fork) — to Mixen: pay from your balance in roubles, with streaming and tool calling over our OpenAI-compatible /v1/chat/completions.

Installation

npm install -g @qwen-code/qwen-code@latest

Node.js 22+ is required. Alternatives — Homebrew (brew install qwen-code) and standalone scripts for Linux/macOS and Windows in the official repository.

Configuration

Qwen Code speaks several protocols (openai, anthropic, gemini, vertex-ai), and Mixen plugs in as a provider on the openai protocol — the CLI supports any OpenAI-compatible endpoint, with each model entry carrying its own baseUrl.

Create ~/.qwen/settings.json (user scope is the recommended place for modelProviders):

{
  "security": {
    "auth": { "selectedType": "openai" }
  },
  "modelProviders": {
    "openai": [
      {
        "id": "openai/gpt-5.6-sol",
        "name": "GPT-5.6 Sol (Mixen)",
        "envKey": "MIXEN_API_KEY",
        "baseUrl": "https://api.mixen.ai/v1"
      },
      {
        "id": "z-ai/glm-5.3-flash",
        "name": "GLM 5.3 Flash (Mixen)",
        "envKey": "MIXEN_API_KEY",
        "baseUrl": "https://api.mixen.ai/v1"
      }
    ]
  },
  "model": { "name": "openai/gpt-5.6-sol" }
}

Key points:

• security.auth.selectedType: "openai" — which protocol to use at startup; without it the CLI stays on its default provider and never reaches Mixen.
• id is the Mixen model ID, sent to the API as is; model.name must match one of the ids in the list.
• envKey is the name of the environment variable holding the key, not the key itself. If omitted, the openai protocol falls back to OPENAI_API_KEY — but then a real OpenAI key on the same machine gets in the way, so a dedicated name is cleaner.
• Entries are told apart by id + baseUrl — keep as many Mixen models in the list as you need.

The API key environment variable (mxn-… is issued in the dashboard):

export MIXEN_API_KEY="your-api-key"   # add to ~/.bashrc or ~/.zshrc to persist

The CLI looks for .env in this order: .qwen/.env → .env → ~/.qwen/.env → ~/.env (the first file found wins; variables are not merged). Source priority: CLI flags → shell variables → .env → the env block of settings.json.

Running

qwen

Useful session commands: /auth — switch provider/protocol, /model — switch model (the choice persists across sessions), /effort — reasoning level.

A quick trial without touching the config — the two credential flags (the only ones that exist):

qwen --auth-type openai --model z-ai/glm-5.3-flash \
  --openai-api-key your-api-key --openai-base-url https://api.mixen.ai/v1

Verifying the connection

The key and balance are checked with a minimal generating request:

curl https://api.mixen.ai/v1/chat/completions \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"z-ai/glm-5.3-flash","messages":[{"role":"user","content":"ping"}],"max_tokens":5}'

200 — the key and balance work; 401 — the key was copied incompletely or it is not a Mixen API key; 402 — the balance is empty. GET /v1/models does not check the key — it answers without one.

Picking a model per task

Every agent turn resends the accumulated context, so input price matters more than it seems. Prices are per 1M tokens, input/output; full tables with cache columns — in the catalog.

Task	Model	Input / output
Code questions, review, small edits	GLM 5.3 Flash (z-ai/glm-5.3-flash)	8.4 / 27.9 ₽
Multi-file edits in familiar code	GLM 5.3 (z-ai/glm-5.3)	156 / 490.2 ₽
Complex agentic tasks, unfamiliar codebase	GPT-5.6 Sol (openai/gpt-5.6-sol)	213.9 / 1069.5 ₽

All three are reasoning models: thinking works without extra setup; tool calling support for a specific model is visible in its catalog card.

Tips and specifics

• Cache savings. Repeated context within 5 minutes is billed at ~10% of the input price (21.4 ₽ per 1M for gpt-5.6-sol instead of 213.9 ₽, 1.7 ₽ for glm-5.3-flash), and every hit extends the window by another 5 minutes — a dense session stays cached from start to finish. Don’t edit project rules (QWEN.md, AGENTS.md) mid-session: changing the prefix resets the cache.
• The limit is 60 RPM per key. That is plenty for dense agent work; on 429 errors reduce the parallelism of subtasks in the client.
• Reasoning eats the response budget — on simple tasks it pays to dial it down.

Troubleshooting

• 401 / Invalid API Key — MIXEN_API_KEY is not set in the session running the CLI (check echo $MIXEN_API_KEY), or the key was copied incompletely.
• “Model not found” — model.name must match the id of a modelProviders entry, not its name field.
• Requests go somewhere other than Mixen — security.auth.selectedType is missing (check /auth), or /model has a different provider’s model selected.
• Variables from .env are not picked up — only the first .env found in the search chain applies; variables from different files are not merged.