QwenCloud

​

Setup

from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.ngrok.ai/v1",
    api_key="ng-xxxxx-g1-xxxxx",
)

response = client.chat.completions.create(
    model="qwencloud:qwen/qwen3.8-max",
    messages=[{"role": "user", "content": "Hello!"}]
)

​

Thinking and tool calls

• Thinking: QwenCloud turns thinking on and off with its own enable_thinking field, and sets a token budget with thinking_budget. The gateway forwards both. It removes reasoning_effort and lists it in the Ngrok-AIG-Removed-Params response header. Qwen3.8 Max thinks by default. Some models, such as z-ai/glm-5.3 and qwen/qwen3.8-2.4t-a95b, always think, and QwenCloud rejects requests that set enable_thinking to false. The gateway also removes both fields for the Qwen3.5 Omni models, because QwenCloud refuses thinking on their audio requests.
• Tool choice: QwenCloud doesn’t support tool_choice: "required". While thinking is on, tool_choice accepts only auto or none, so turn thinking off to force a specific tool. The gateway forwards tool_choice as sent, so QwenCloud returns the error.

response = client.chat.completions.create(
    model="qwencloud:qwen/qwen3.8-max",
    messages=[{"role": "user", "content": "Hello!"}],
    extra_body={
        "enable_thinking": True,
        "thinking_budget": 4096,
    },
)

​

Available models

​

Next steps

• Provider keys
• Access key configurations