ZEN gemini-3.6-flash returns "Upstream request failed" while gemini-3.5-flash works

[gozimi]

Description

gemini-3.6-flash consistently returns an upstream error, while gemini-3.5-flash works with the exact same request.

Plugins

No response

OpenCode version

No response

Steps to reproduce

Request

curl -X POST \
  "https://opencode.ai/zen/v1/models/gemini-3.6-flash:generateContent" \
  -H "x-goog-api-key: <ZEN_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "parts": [
          {
            "text": "Hello!"
          }
        ]
      }
    ]
  }'

**Result**
{
  "error": {
    "code": 400,
    "message": "Error from provider (Console): Upstream request failed",
    "status": "INVALID_ARGUMENT"
  }
}

**Expected**
The request should succeed and return a normal Gemini response.

**Additional information**
GET /v1/models includes gemini-3.6-flash.
Replacing the model with gemini-3.5-flash (same endpoint, same API key, same payload) works correctly.
Calling the official Google Gemini API directly with gemini-3.6-flash and the same payload also works.

### Screenshot and/or share link

_No response_

### Operating System

_No response_

### Terminal

_No response_

[github-actions]

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

• 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
• Zen returns 400 Upstream request failed when thinking is enabled for latest Claude models (fable-5, sonnet-5, opus-4-8); older models (opus-4-6, haiku-4-5) work fine #35124: Zen returns 400 Upstream request failed for latest Claude models (similar upstream failure pattern)
• OpenCode Go: kimi-k3, kimi-k2.6, kimi-k2.7-code, glm-5.2, grok-4.5, qwen3.7-max, hy3-preview all fail with 'Upstream request failed' — non-standard 'mcp'/'system' fields rejected by strict provider validators #37771: Multiple Zen/Go models fail with 'Upstream request failed' — upstream provider rejects non-standard fields

If your issue is specifically about gemini-3.6-flash being listed in GET /v1/models but not actually working (unlike gemini-3.5-flash), it may be a distinct model availability bug rather than a general upstream failure — in that case, please clarify that in your report so maintainers can triage it separately.

[kai-kleinbard-tyto]

Closing — insufficient information to investigate (no template, no version, no repro steps, or support question). Please reopen with the Bug Report template.