OpenCode 服务故障反馈(go 套餐用尽后，无法转到zen 余额) #39470

Go 套餐 /zen/go/v1/chat/completions 全模型返回 HTTP 500

Bug Description

Go 套餐周用量耗尽后，我充值了 Zen 余额并开启了"达到使用限额后使用可用余额"。但 /zen/go/v1/chat/completions 端点开始对所有模型返回 HTTP 500，包括 glm-5.2、deepseek-v4-pro、kimi-k3 等。这看起来是服务端故障，不是余额或账户问题（/v1/models 返回 200，Zen 余额显示正常）。

Environment

OpenCode version: (claude-g5.2-go via local liteLLM gateway)
Operating System: macOS 26.5.2
Terminal: Terminal.app
Network: Clash Party TUN proxy (also tested with direct curl, same result)

Steps to Reproduce

Go 套餐周用量耗尽
充值 Zen 余额，开启"达到限额后使用可用余额"
调用 /zen/go/v1/chat/completions 任意模型
返回 HTTP 500

Actual Behavior

curl -X POST https://opencode.ai/zen/go/v1/chat/completions
-H "Authorization: Bearer "
-H "Content-Type: application/json"
-d '{"model": "glm-5.2", "messages": [{"role": "user", "content": "hi"}]}'

Response:
HTTP/2 500
{"type":"error","error":{"message":"Internal server error"}}

Tested models: glm-5.2, deepseek-v4-pro, kimi-k3, kimi-k2.6, qwen3.7-max, minimax-m3 — all return 500.

Expected Behavior

Chat completions should return a normal response, or at least a clear error (e.g., 429 rate limit, 402 insufficient balance) rather than an opaque 500.

Additional Context

OpenCode account: zuwei2046@126.com
Workspace ID: wrk_01KSD5TPCHY8XBHHPMQ64PE261
GET /zen/go/v1/models returns HTTP 200 with all 23 models listed correctly.
Zen balance shows as active in the dashboard.
Direct curl (no proxy, no gateway) produces the same 500, so this is not a local network or gateway issue.
This started around ~14:00 CST on 2026-07-29 and has persisted for hours.

GET /zen/go/v1/models returns HTTP 200 with all 23 models listed correctly.
Zen balance shows as active in the dashboard.
Direct curl (no proxy, no gateway) produces the same 500, so this is not a local network or gateway issue.
This started around ~14:00 CST on 2026-07-29 and has persisted for hours.

Note: I initially thought this might be a billing/quota fallback issue, but the fact that /v1/models works fine and the error is a generic 500 across all models suggests a server-side outage rather than an account problem.