Zen: gpt-5.6-luna / gpt-5.6-terra fail with "Error from provider (Console): Upstream request failed"

[geminaldiol2333-commits]

Zen: gpt-5.6-luna / gpt-5.6-terra fail with "Error from provider (Console): Upstream request failed"

Description

Using Zen (opencode provider), gpt-5.6-luna and gpt-5.6-terra consistently fail with HTTP 403 Error from provider (Console): Upstream request failed, while gpt-5.4-nano, gemini-3.6-flash and claude-sonnet-5 work fine with the same API key. Both models are listed in GET https://opencode.ai/zen/v1/models, so access appears enabled. The error string matches #39293 (gemini-3.6-flash), suggesting the "Console" upstream is failing for these models.

Plugins

None

OpenCode version

1.18.9

Steps to reproduce

1. curl -X POST "https://opencode.ai/zen/v1/responses" -H "Authorization: Bearer <ZEN_KEY>" -H "Content-Type: application/json" -d '{"model":"gpt-5.6-luna","input":"hi"}'
   → HTTP 403: {"model":"gpt-5.6-luna","error":{"param":null,"message":"Error from provider (Console): Upstream request failed"}}
2. Same request with gpt-5.6-terra → same HTTP 403 error
3. Same request with gpt-5.4-nano → HTTP 200, normal response

Screenshot and/or share link

No response

Operating System

Windows 11

Terminal

Windows Terminal

[github-actions]

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

• OpenCode Zen lists GPT 5.6 Terra but cannot run it #38624: OpenCode Zen lists GPT 5.6 Terra but cannot run it (same model, same Zen provider failure)
• All paid OpenCode Zen models fail with 'Upstream request failed' — free models work #36506: All paid OpenCode Zen models fail with 'Upstream request failed' — free models work (same error pattern across Zen paid models)
• ZEN gemini-3.6-flash returns "Upstream request failed" while gemini-3.5-flash works #39293: ZEN gemini-3.6-flash returns "Upstream request failed" (same error string you referenced, different model)

[ltceu6o5fu]

I am also experiencing the same issue. When making an API request using the curl command, the server returns an invalid_prompt error indicating an upstream request failure.

To Reproduce
Send a request to the API using curl.
Target the gpt-5.6-luna model.
Expected behavior
The request should be processed successfully and return a valid response.

Actual behavior
The request(https://opencode.ai/zen/go/v1/responses) fails and returns the following JSON error message:

{"model":"gpt-5.6-luna","error":{"code":"invalid_prompt","message":"Error from provider (Console Go): Upstream request failed"}}

[xxxxchaos]

Reproduced from a Hong Kong production server via the documented Zen API endpoint. Additional comparison details from #40162:

• gpt-5.6-luna: HTTP 403
• gpt-5.6-terra: HTTP 403
• gpt-5.4-mini: HTTP 200 with the same Zen API key and host
• claude-haiku-4-5: HTTP 200 with the same Zen API key and host
• Both IPv4 and IPv6 reproduce the GPT-5.6 failure

The more specific upstream error is:

Error from provider (Console): Upstream request failed: [unsupported_country_region_territory] Country, region, or territory not supported

This model-level comparison suggests the Zen credential and general connectivity are valid, and the failure is specific to the current GPT-5.6 upstream route rather than all Zen traffic from the host.