Qwen3.5 397B A17B

Summary: Qwen3.5 397B A17B is a large-scale multimodal Mixture-of-Experts (MoE) model from Alibaba Cloud's Qwen team, combining a hybrid linear-attention and sparse-MoE architecture with 397B total parameters and 17B active parameters per token. The model natively supports an extended 262k token context window, reasoning ("thinking mode"), tool calling, and multilingual understanding across 201 languages and dialects. In addition to text, it processes image input, making it suitable for advanced conversational agents, complex reasoning tasks, and vision-language applications at scale.

Intelligence	Speed	Input	Output
High	Low	Text, Image	Text

Central parameters

Description: Large-scale MoE model from Alibaba Cloud with 397B total parameters and 17B active parameters per token, combining hybrid linear attention with sparse expert routing for efficient inference at scale.

Model identifier: Qwen/Qwen3.5-397B-A17B

IONOS CLOUD AI Model Hub Lifecycle and Alternatives

IONOS CLOUD start date	End of Life	Alternative	Successor
July 14, 2026	N/A

Origin

Provider	Country	License	Flavor	Release
Alibaba Cloud	China	Apache 2.0	Instruct	February 2026

Technology

Context window	Parameters	Quantization	Multilingual	Further details
262k	397B (17B active)	FP8	Yes	Hugging Face

Modalities

Text	Image	Audio
Input and output	Input	Not supported

Endpoints

Chat completions	Embeddings	Image generation
v1/chat/completions	Not supported	Not supported

Features

Streaming	Reasoning	Tool calling
Supported	Supported	Supported

Reasoning example

Qwen3.5 397B A17B operates in thinking mode by default, generating an internal reasoning trace before producing its final response. It makes the model well-suited for complex, multi-step reasoning tasks.

Unlike models that offer configurable reasoning effort levels (low, medium, high), Qwen3.5 397B A17B does not support this granularity. Instead, reasoning is either on (default) or off. To deactivate it, set "reasoning_effort": "none".

Note: The reasoning trace counts toward max_tokens. If max_tokens is set too low, reasoning can consume the entire token budget, leaving no tokens for the final answer. Activating reasoning also increases response latency, since the model generates the reasoning trace in addition to the final answer.

Request

{
  "stream": false,
  "model": "Qwen/Qwen3.5-397B-A17B",
  "messages": [
    {
      "role": "user",
      "content": "Answer me with one letter. Maybe A."
    }
  ]
}

Response (shortened for readability)

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "A",
        "reasoning": "User asks to answer with one letter, maybe A. So respond with a single letter."
      }
    }
  ]
}

Deactivate reasoning

To skip the internal reasoning trace and return only the final answer, set "reasoning_effort": "none".

{
  "stream": false,
  "model": "Qwen/Qwen3.5-397B-A17B",
  "reasoning_effort": "none",
  "messages": [
    {
      "role": "user",
      "content": "Answer me with one letter. Maybe A."
    }
  ]
}

Usage example

Chat completions

The following example demonstrates how to use Qwen3.5 397B A17B for complex reasoning tasks, such as analyzing data trends and making predictions.

API Endpoint: POST https://openai.inference.de-txl.ionos.com/v1/chat/completions

Request:

{
  "model": "Qwen/Qwen3.5-397B-A17B",
  "messages": [
    {
      "role": "system",
      "content": "You are an expert data analyst. Analyze the provided data and give a summary of the key trends."
    },
    {
      "role": "user",
      "content": "Here is the sales data for Q1: Jan: $10k, Feb: $12k, Mar: $15k. Predict the trend for Q2 based on this growth."
    }
  ],
  "temperature": 0.5,
  "max_tokens": 1000
}

Response:

{
  "id": "chatcmpl-890",
  "object": "chat.completion",
  "created": 1677652289,
  "model": "Qwen/Qwen3.5-397B-A17B",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "Based on the Q1 data, there is a consistent monthly growth of roughly 20-25%. If this trend continues, Q2 sales are projected to reach approximately $18k in April, $21.6k in May, and $26k in June.",
      "reasoning": "The user provided Q1 sales data (Jan: 10k, Feb: 12k, Mar: 15k). Calculating growth: Jan to Feb is +20%, Feb to Mar is +25%. I will assume a continued growth trend of roughly 20-25% for Q2 (April, May, June) to make the prediction.",
      "reasoning_content": "The user provided Q1 sales data (Jan: 10k, Feb: 12k, Mar: 15k). Calculating growth: Jan to Feb is +20%, Feb to Mar is +25%. I will assume a continued growth trend of roughly 20-25% for Q2 (April, May, June) to make the prediction."
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 45,
    "completion_tokens": 48,
    "total_tokens": 93
  }
}

Vision model example

{
  "model": "Qwen/Qwen3.5-397B-A17B",
   "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "What do you see in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://www.ionos.com/favicon.ico?v2",
              "detail": "auto"
            }
          }
        ]
      }
    ],
    "temperature": 0.7,
    "max_tokens": 1000
  }

Stream chat completions

To receive responses in real time, you can use streaming. When streaming is selected, usage statistics are not included by default. To obtain usage information in the final stream chunk, you must explicitly set "stream_options": {"include_usage": true}.

API Endpoint: POST https://openai.inference.de-txl.ionos.com/v1/chat/completions

Request:

{
  "model": "Qwen/Qwen3.5-397B-A17B",
  "messages": [
    {
      "role": "user",
      "content": "Compose a poem about the sea."
    }
  ],
  "stream": true,
  "stream_options": {
    "include_usage": true
  }
}

Response:

...
data: {"choices":[],"created":1769419547,"id":"chatcmpl-3d0323bb-67f8-41c5-b757-8fc5a48230f4","model":"Qwen/Qwen3.5-397B-A17B","object":"chat.completion.chunk","usage":{"completion_tokens":26,"prompt_tokens":78,"total_tokens":104}}

data: [DONE]

Rate limits

Rate limits ensure fair usage and reliable access to the AI Model Hub. In addition to the contract-wide rate limits, no model-specific limits apply.