Bug: /v1/messages streaming for qwen3.6-plus lacks content_block_stop/message_delta/message_stop, causing Claude Code to hang

⚠️ 提交前确认

• 我已确认当前使用的是最新版本的 New API
  当前版本
  v1.0.0-rc.4
• 我已搜索过现有 issues，未找到完全相同的问题
• 我已提供最小复现请求
• 我已隐藏所有敏感信息，例如 API Key、数据库密码、Token

版本信息

New API 部署方式：Docker

镜像：

calciumion/new-api:latest


复现步骤

执行以下请求：

curl -N http://YOUR_NEW_API_HOST:8979/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk-REDACTED" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "qwen3.6-plus",
    "max_tokens": 4096,
    "stream": true,
    "extra_body": {
      "enable_thinking": false
    },
    "messages": [
      {
        "role": "user",
        "content": "你好，请只回复一句话。"
      }
    ]
  }'
  
  实际行为

实际返回的 SSE 流大致如下：

event: message_start
data: {"type":"message_start", ...}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"..."}}

...

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"你好，"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"我已按你的要求"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"只回复"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"这一句话。"}}

# 然后直接 EOF，没有后续事件

缺少以下事件：

event: content_block_stop
data: {"type":"content_block_stop","index":1}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"}, ...}

event: message_stop
data: {"type":"message_stop"}

完整日志中可以看到 qwen3.6-plus 已经输出了 text block 和 text_delta，但是没有正常输出 content_block_stop、message_delta、message_stop。


期望行为

当上游模型流式输出结束时，New API 的 Anthropic /v1/messages 兼容层应该补齐完整的 Anthropic SSE 结束事件。

期望流式响应类似：

event: message_start
event: content_block_start
event: content_block_delta
event: content_block_stop
event: content_block_start
event: content_block_delta
event: content_block_stop
event: message_delta
event: message_stop

即至少在最后包含：

event: content_block_stop
data: {"type":"content_block_stop","index":1}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"}}

event: message_stop
data: {"type":"message_stop"}
对比测试

同样请求更换模型为 mimo-v2.5-pro，流式响应可以正常结束：

event: content_block_stop
data: {"type":"content_block_stop","index":1}

event: message_delta
data: {"type":"message_delta","usage":{...},"delta":{"stop_reason":"end_turn"}}

event: message_stop
data: {"type":"message_stop"}

因此问题看起来不是 Claude Code 或 curl 客户端问题，而是 qwen3.6-plus 通过 New API /v1/messages streaming 转换时，SSE 结束事件未被正确输出。

额外测试

尝试在请求体中添加：

{
  "enable_thinking": false
}

以及：

{
  "extra_body": {
    "enable_thinking": false
  }
}

但返回中仍然包含大量：

{
  "type": "thinking_delta"
}

说明 /v1/messages 路径下该参数可能没有被正确透传，或者当前渠道不支持该参数位置。

不过即使 thinking 没关闭，理论上只要最终 text block 后补齐 content_block_stop、message_delta 和 message_stop，Claude Code 也可以继续处理。

New API 最新版下，qwen3.6-plus 通过 /v1/messages stream=true 已能输出 text_delta，但缺少最终的 content_block_stop、message_delta、message_stop，导致 Claude Code 等不到结束信号而卡住

Start at the /v1/messages Anthropic streaming conversion entry point and reproduce the qwen3.6-plus request with the provided curl command. Trace the stream through the final text_delta and compare its termination with the mimo-v2.5-pro response. Done means the qwen3.6-plus stream emits content_block_stop, message_delta, and message_stop before EOF, with Claude Code no longer hanging.