# [BUG] Most Qwen AI (International) models return an empty stream (`data: [DONE]`) while only Qwen3.6-27B and Qwen3-Coder work

[tacitvskilgore]

Description

The Qwen AI (International) provider appears to be broken for most currently available models.

Only the following models work correctly:

• Qwen3.6-27B
• Qwen3-Coder

Almost every other Qwen AI (International) model returns an empty response.

This is not a Cherry Studio issue. The problem can be reproduced by calling Chat2API's OpenAI-compatible endpoint directly.

---

Environment

• Chat2API: v1.4.0

• OS: Windows 10 x64

• Provider: Qwen AI (International)

• Authentication: JWT login

• Endpoint:

  POST /v1/chat/completions
  

• Client tested:

  • Cherry Studio
  • PowerShell (direct HTTP requests)

---

Models Tested

✅ Working

• Qwen3.6-27B
• Qwen3-Coder

❌ Broken

• Qwen3.7-Plus
• Qwen3.7-Max
• Qwen3.6-Plus
• Qwen3.6-Max-Preview
• Qwen3.6-35B-A3B
• Qwen3.5-Plus
• Qwen3.5-Omni-Plus
• Qwen3.5-Flash
• Qwen3.5-Max-Preview
• Qwen3.6-Plus-Preview
• Qwen3.5-397B-A17B
• Qwen3.5-122B-A10B
• Qwen3.5-Omni-Flash
• Qwen3.5-27B
• Qwen3.5-35B-A3B
• Qwen3-Max
• Qwen3-235B-A22B-2507
• Qwen3-VL-235B-A22B
• Qwen3-Omni-Flash
• Qwen3-Omni

---

Steps to Reproduce

Request:

POST /v1/chat/completions

Body:

{
  "model": "Qwen3.7-Max",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ],
  "stream": true
}

---

Actual Result

The server immediately returns:

data: [DONE]

No streaming chunks are produced.

---

With stream=false:

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "",
        "reasoning_content": ""
      },
      "finish_reason": "stop"
    }
  ]
}

The request succeeds, but the response is empty.

---

Expected Result

The server should stream normal OpenAI-compatible chunks, for example:

data: { ... delta ... }

data: { ... delta ... }

data: { finish_reason: "stop" }

data: [DONE]

---

Comparison

Qwen3.6-27B (Working)

Produces normal SSE chunks:

data: {"choices":[{"delta":{"role":"assistant","content":""}}]}

data: {"choices":[{"delta":{"content":"Hello"}}]}

...

data: {"choices":[{"finish_reason":"stop"}]}

data: [DONE]

Qwen3.7-Max (Broken)

Produces only:

data: [DONE]

No delta events are sent.

---

Additional Observation

Even the two working models (Qwen3.6-27B and Qwen3-Coder) do not appear to preserve conversation context correctly.

Each request behaves like a completely new conversation, even when previous messages are included.

I'm not sure whether this is related to the same provider implementation, but it may indicate another issue with the Qwen AI (International) provider.

---

Notes

This issue seems related to #166, but the reproduction here is more specific.

The provider successfully returns HTTP 200, but no assistant content is ever emitted for most Qwen AI (International) models.

It looks like the provider may not be parsing the newer streaming response format used by these models.