OpenCode Go: "Upstream request failed: Endpoint is unavailable" for qwen/grok models (server-side, not client network)

Description

Summary

OpenCode Go (Zen gateway) returns server_error: Error from provider (Console Go): Upstream request failed: Endpoint is unavailable. for qwen3.7-plus / qwen3.7-max / grok-4.5, while other Go models (kimi-k3, glm-5.2, minimax-m3, gpt-5.6-luna) work fine with the same API key. This is reproducible via direct curl calls to the gateway — not a client-side network/proxy issue.

Environment

• opencode CLI version: 1.18.11 (Windows 10, win32-x64)
• Auth: opencode-go provider, type "api" (Go subscription, bound via /connect)
• Default model: qwen3.7-plus

Reproduction (curl, no SDK involved)

POST https://opencode.ai/zen/go/v1/chat/completions
Authorization: Bearer
Content-Type: application/json

{"model":"qwen3.7-plus","messages":[{"role":"user","content":"hi"}],"stream":false,"max_tokens":16}

Actual response

{"error":{"type":"server_error","message":"Error from provider (Console Go): Upstream request failed: Endpoint is unavailable."}}

Observations

• Same error with and without HTTP proxy; gateway is reachable from here (HTTP 200 on /zen/go/v1/models).
• Model matrix with the same key:
  • WORKS: kimi-k3, glm-5.2, minimax-m3, gpt-5.6-luna
  • FAILS: qwen3.7-plus, qwen3.7-max, grok-4.5 -> "Upstream request failed: Endpoint is unavailable."
• The error string is NOT present in the opencode client binary, so it originates from the gateway's upstream provider ("Console Go").
• opencode log (opencode.log) shows: AI_APICallError: Upstream request failed: Endpoint is unavailable. with providerID=opencode-go modelID=qwen3.7-plus, every request, 100% failure rate.

Expected

Chat completions succeed for all models listed in the Go model catalog.

Severity

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

Start by reproducing the POST request against https://opencode.ai/zen/go/v1/chat/completions with the listed failing and working models, then compare the opencode.log providerID and modelID entries. Done means qwen3.7-plus, qwen3.7-max, and grok-4.5 return successful chat completions through the Go gateway rather than the upstream endpoint error.