gpt-5.6-luna 经 OpenCode Go 中继访问返回 403: "This model is not available in your region" #41518

Activity

1. github-actions

   assigned

   fwang

   on Aug 10, 2026
2. github-actions commented on Aug 10, 2026

   

   github-actionsBot

   on Aug 10, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • [bug] GPT-5.6 Luna/Terra return unsupported_country_region_territory via Zen from Hong Kong #40162: GPT-5.6 Luna/Terra return unsupported_country_region_territory via Zen from Hong Kong (same model, same 403 region error)
   • Zen: gpt-5.6-luna / gpt-5.6-terra fail with "Error from provider (Console): Upstream request failed" #39831: Zen: gpt-5.6-luna / gpt-5.6-terra fail with "Error from provider (Console): Upstream request failed" (same model availability issue)
   • I have subscribed to GO, but I found that the GO subscription is basically unusable. #40791: GO subscription models unusable (related GO relay access issues)
   • Console Go (zen/go/v1): deepseek-v4-flash returns RegionError requiring China opt-in, no programmatic opt-in path #39872: Console Go (zen/go/v1): deepseek-v4-flash returns RegionError requiring China opt-in (similar regional restriction pattern)
3. wulishuhan commented on Aug 10, 2026

   

   wulishuhan

   on Aug 10, 2026

   More actions

   Same issue here, with additional evidence that may help isolate the cause.

   Setup: Codex → opencodex local proxy → OpenCode Go relay (https://opencode.ai/zen/go/v1) → gpt-5.6-luna.

   Findings:

   • The API key is valid: same key returns HTTP 200 for deepseek-v4-flash and kimi-k2.7-code, and the provider connection test succeeds (25 models discoverable, including gpt-5.6-luna).
   • The egress region is NOT the problem: I tested the actual proxy egress used by the relay calls via https://opencode.ai/cdn-cgi/trace and it reports loc=SG (Singapore, 18.143.67.40). I also previously hit the same 403 while the connection exited via a US node (cf-ray ...-EWR). Both Singapore and US are OpenAI-supported regions, yet gpt-5.6-luna still returns:
     Error from provider (Console Go): Upstream request failed: [403] This model is not available in your region.
   • The account was registered from mainland China.

   Questions:

   1. Is the region check based on the account's registered country rather than the request egress IP?
   2. If so, can GPT models be enabled for accounts registered in mainland China, or is this out of scope for the Go plan?
   3. The Go plan page advertises gpt-5.6-luna as included — if mainland-China accounts cannot use it, what is the remedy (enablement or refund)?

   cc @fwang
4. q531620267 commented on Aug 14, 2026

   

   q531620267

   on Aug 14, 2026

   More actions

   @fwang
   Mainland China-registered account cannot use gpt-5.6-luna — request confirmation of region restriction basis and remedy

   I'm experiencing the same issue, and I'd like to share my verified findings for the maintainers' reference.

   Environment

   Account region: Mainland China
   Plan: OpenCode Go subscription (gpt-5.6-luna is explicitly listed on the plan page)
   Access method: via an OpenAI-compatible client connecting directly to https://opencode.ai/zen/go/v1
   Verification results (with API key)

   GET /v1/models → 200, gpt-5.6-luna is indeed present in the model list
   Same key with deepseek-v4-pro and kimi-k3 → 200, works fine
   POST /v1/responses (model: gpt-5.6-luna) → 403, returning: "Error from provider (Console Go): Upstream request failed: [403] This model is not available in your region."
   Variables already ruled out

   Egress IP is not the cause: my egress is China Telecom (mainland). Also, another user in this thread tested Singapore (SG) and US (EWR) nodes and still got 403.
   This is unrelated to the "enable models deployed in China" opt-in toggle: that toggle affects the RegionError for deepseek-v4-flash; luna is subject to a different restriction.
   Questions

   Is the region check based on the account registration country, rather than the request egress IP?
   If so, the Go plan advertises gpt-5.6-luna, yet accounts registered in Mainland China (equally paying customers, and presumably within the supported country list) cannot use it at all. Are there any remedies available, such as account migration, manual approval, or a refund?
   Is there an official list of supported regions for this model? Are mainland China accounts planned to be supported in the future?
   Thanks in advance for any official clarification.
5. SangYu666 commented on Aug 14, 2026

   

   SangYu666

   on Aug 14, 2026

   More actions

   代理开TUN模式，亲测有用

   👍React with 👍7Reacted by tclxxiaofeng-wang, yayozen, kolanioo, dongmingre, HydrogenDelta, 7214978 and xlu137492
6. q531620267 commented on Aug 14, 2026

   

   q531620267

   on Aug 14, 2026 via email

   More actions

   谢谢，我试试~ Øimmortal. ***@***.***
7. q531620267 commented on Aug 14, 2026

   

   q531620267

   on Aug 14, 2026 via email

   More actions

   确实有用，太感谢~~ Øimmortal. ***@***.***
8. imkawa commented on Aug 18, 2026

   

   imkawa

   on Aug 18, 2026

   Last edited by imkawa

   More actions

   Issue Description
   When using OpenCode to call GPT-5.6-luna, intermittent failures occur:

   Sometimes it responds normally;
   Other times it returns the error: Upstream request failed: [403] This model is not available in your region.
   Troubleshooting Process
   Checked router rule matching: Found that OpenCode requests sometimes fail to match any proxy/routing rules, making it impossible to determine whether the traffic actually went through the proxy.
   Checked network connection method: Discovered that OpenCode establishes connections using IPv4 + IPv6 dual-stack.
   Identified the cause of rule failure: The proxy/routing rules on the router cannot match or hijack IPv6 traffic. This portion of traffic exits directly using the real IP (direct connection).
   Root Cause Analysis
   When OpenCode connects via IPv4, traffic is captured by router rules and forced through the proxy, so it can be accessed normally.

   When OpenCode connects via IPv6, traffic bypasses proxy rules and exits via direct connection. The server detects that the real IP's region is restricted and returns a 403 region restriction error.

   Solution
   Disable IPv6 on the computer to force OpenCode to establish connections only via IPv4.

   This ensures all traffic is handled by the router's proxy rules, and the 403 region-unavailable issue caused by IPv6 direct connection no longer occurs.

   Summary
   IPv6 traffic bypasses router proxy rules → Real IP is exposed → Triggers region restriction 403.

   Disabling client IPv6 and forcing pure IPv4 connections unifies all traffic through the proxy, which completely resolves this issue.
9. zhaozhiguoya commented on Aug 24, 2026

   

   zhaozhiguoya

   on Aug 24, 2026

   More actions

   补充一组复现数据，可以进一步缩小范围（Windows 11，OpenCode Desktop，provider: opencode-go）：

   复现特征

   1. 错误随时间变化

   • 2026-08-22 全天：Upstream request failed: [server_error] Upstream response was not valid JSON + Service Unavailable（约 17 次全部失败）
   • 2026-08-24 起：变为 [unsupported_country_region_territory] Country, region, or territory not supported

   2. 同一时刻"小请求成功、大请求失败"（关键发现）

   2026-08-24 ~00:49 UTC，同一 key、同一出口 IP：

   请求方式	结果
   App 内旧会话（大上下文，ses_fd748f54affeLoVxYRt7qiDX4Q）	❌ unsupported_country_region_territory
   直接 curl /zen/go/v1/responses 裸请求 "hi"	✅ completed
   同上 + tools 定义	✅ completed
   同上 + ~40KB 输入	✅ completed
   同上 + "stream": true	✅ HTTP 200 SSE 正常
   同上 + "store": true	✅ completed

   连续 6 次裸请求 100% 成功的同时，App 内请求持续失败。

   推测

   结合 #40180 的分析：网关不同节点对客户端地理头（cf-ipcountry 等）的处理不一致——
   部分节点剥离了头所以成功，部分节点原样转发导致被上游 OpenAI 拒绝。
   负载均衡落在哪个节点决定了成败，表现为"时好时坏"。
   另外大上下文会话可能被路由到固定/特定节点池，导致旧会话必现失败而新请求随机成功。

   环境

   • OpenCode Desktop（@opencode-aidesktop），Windows 11
   • provider: opencode-go，model: gpt-5.6-luna
   • 同 key 下 glm-5.3 / deepseek-v4-pro / big-pickle 均正常
10. thdxr commented on Oct 11, 2026

    

    thdxr

    on Oct 11, 2026

    Member

    More actions

    Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

    — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
11. thdxr

    closed this as not plannedon Oct 11, 2026