DeepSeek & ZEN models: request timeout with no response on Windows 11 25H2 (Node.js 24.12.0, opencode v1.17.15) #36012

Description

Description
When using opencode with any model – whether a self-configured DeepSeek provider or the built-in ZEN plan model – the request always times out after ~20 seconds. The session is created and the stream event is logged, but no data or error is ever received from the API. The only output is build · and then idle until abort/resume.

This occurs even with the official ZEN model, which should work out-of-the-box without any configuration.

Steps to Reproduce

1.Fresh installation of opencode v1.17.15 on Windows 11 25H2.

2.(Optional) Configure opencode.json for DeepSeek (see config below).

3.Run:

powershell
opencode run "hello" --print-logs --log-level DEBUG
(or --model zen, --model deepseek/deepseek-chat, etc.)

opencode_debug.log

Session aborted.

Expected behavior
AI should respond with a message (either streaming or full response).

Actual behavior
Complete silence until timeout. No network error, no HTTP status code, no data chunk.

Environment

OS: Windows 11 Pro 25H2 (build 26100.XXXX)

Node.js: v24.12.0

opencode: v1.17.15 (installed globally via npm)

Shell: PowerShell 7.4.x (also tested in CMD)

Network: Direct internet access, no corporate proxy (also tested with HTTP_PROXY cleared)

Configuration used (for DeepSeek, but ZEN also fails)

json
{
"providers": {
"deepseek": {
"type": "openai",
"apiKey": "sk-...",
"baseURL": "https://api.deepseek.com/v1",
"models": {
"deepseek-chat": { "name": "DeepSeek Chat" }
}
}
}
}
Additional context

curl directly to the DeepSeek API (deepseek-chat) works perfectly, so the API key and network are functional.

The issue persists even when no configuration file is present (using default ZEN model).

Tried:

--thinking on/off

--model deepseek/deepseek-chat and deepseek/deepseek-v4-flash

Adding "type": "openai"

Setting "streaming": false in config (no effect)

Running in CMD instead of PowerShell

Deleting and recreating ~/.config/opencode

Reinstalling opencode globally

No HTTP logs appear even with NODE_DEBUG=http,https (possibly swallowed by opencode).

The error is consistent across all models.

Log snippet (typical)

text
timestamp=... level=INFO run=... message=stream providerID=deepseek modelID=deepseek-chat session.id=... small=true agent=title mode=primary
timestamp=... level=INFO run=... message="llm runtime selected" llm.runtime=ai-sdk llm.provider=deepseek llm.model=deepseek-chat
timestamp=... level=DEBUG run=... message="Discovered session ..."
... (many Discovered session lines)
timestamp=... level=INFO run=... message="Parent ... stuck with no active subagents. Triggering abort+resume."
timestamp=... level=INFO run=... message="Abort+Resume on ... (20s idle). Aborting..."
Possible cause
The underlying ai-sdk may be constructing requests that DeepSeek (or the ZEN endpoint) does not accept, leading to a silent hang. Alternatively, there might be a TLS or HTTP/2 compatibility issue with Node.js 24 on Windows 11 25H2.

Request
Please investigate why opencode fails to establish a valid data stream with these models on this environment. I am happy to provide further logs or test patches.

Plugins

opencode-auto-resume

OpenCode version

1.17.15

Steps to reproduce

1.Fresh installation of opencode v1.17.15 on Windows 11 25H2.
2.2.(Optional) Configure opencode.json for DeepSeek (see config below).
3.Run:
powershell
opencode run "hello" --print-logs --log-level DEBUG
(or --model zen, --model deepseek/deepseek-chat, etc.)
4.

Observe the terminal:

Session created

build · appears

No further output for ~20 seconds

Then Parent ... stuck with no active subagents. Triggering abort+resume.

Screenshot and/or share link

Operating System

Windows 11

Terminal

powershell