谢谢。

这三轮测试（两次 stream=False 不同模型 + 一次 stream=True）全部在约 0.7 秒就给出同样的 pending stream has been canceled 可否在你那边实际测试下，直接发送一个1m左右的pdf给模型，看看结果。已经升级到最新版本，之前是30秒，现在直接不到1秒就完了。
gpt-5.4-high 是当下使用最多的，返回 gpt-5.4-high 在当前账
号池中不可用（未订阅或已被封禁）这个是无法开通吗
3.运行日志
通过 SSE 实时流式接收服务端日志这里面的日志是空的。在vps中获取的部分

pm2 logs windsurf-api
[TAILING] Tailing last 15 lines for [windsurf-api] process (change the value with --lines option)
/root/.pm2/logs/windsurf-api-out.log last 15 lines:
0|windsurf | [INFO] POST /v1/chat/completions
0|windsurf | [INFO] GET /v1/models
0|windsurf | [INFO] POST /auth/login (add account)
0|windsurf | [INFO] GET /auth/accounts (list accounts)
0|windsurf | [INFO] DELETE /auth/accounts/:id (remove account)
0|windsurf | [INFO] Model catalog: 87 cloud models, 9 new entries merged
0|windsurf | [INFO] Chat: model=gemini-3.1-pro-high flow=cascade stream=true attempt=1 account=Glenn Conway ls=42100 turns=2 chars=166
0|windsurf | [INFO] Cascade workspace init complete for LS port=42100
0|windsurf | [INFO] Cascade done {cascadeId=e7ecd3b9 reason=idle_done polls=25 textLen=57 thinkingLen=685 stepCount=3 toolCalls=0 sawActive=true sawText=true lastStatus=1 ms=6599}
0|windsurf | [INFO] Cascade usage: in=1574 out=470 cache_r=0 cache_w=0
0|windsurf | [INFO] Chat: model=gemini-3.1-pro-high flow=cascade attempt=1 account=Kimberly Nelson ls=42100 turns=3 chars=50507
0|windsurf | [INFO] Cascade workspace init complete for LS port=42100
0|windsurf | [INFO] Chat: model=gemini-3.1-pro-high flow=cascade attempt=1 account=Glenn Conway ls=42100 turns=3 chars=48420
0|windsurf | [INFO] Chat: model=gpt-5.4-xhigh flow=cascade attempt=1 account=Kimberly Nelson ls=42100 turns=3 chars=48408
0|windsurf | [INFO] Chat: model=gpt-5.4-xhigh flow=cascade stream=true attempt=1 account=Glenn Conway ls=42100 turns=3 chars=48408

prompt #5

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions