gemini-cli - 💡(How to fix) Fix Vertex AI: flash model router maps to non-existent gemini-3-flash

Description

When using Vertex AI authentication, the model router remaps all flash model requests to gemini-3-flash, which does not exist on Vertex AI. This makes it impossible to use any flash model (including gemini-3.5-flash) through the CLI with Vertex AI.

Steps to Reproduce

1. Configure Gemini CLI for Vertex AI auth ("selectedType": "vertex-ai" in settings.json)
2. Run: gemini -m gemini-3.5-flash -p "hello"
3. Observe the error references gemini-3-flash, not the model you requested

Expected Behavior

The CLI should send gemini-3.5-flash to the Vertex AI API (which works — verified via direct curl to the Vertex AI endpoint).

Actual Behavior

The agent router overrides the model to gemini-3-flash regardless of what is passed via -m:

[Routing] Selected model: gemini-3-flash (Source: agent-router/override, Latency: 0ms)

Error:

ModelNotFoundError: Publisher Model `projects/<project>/locations/us-central1/publishers/google/models/gemini-3-flash` was not found

Root Cause

In hasGemini35FlashGAAccess(), the Vertex AI branch sets both flash model slots to gemini-3-flash:

if (authType === AuthType2.USE_GEMINI) {
  setFlashModels("gemini-3-flash-preview", "gemini-3.5-flash");  // correct
} else {
  setFlashModels("gemini-3-flash", "gemini-3-flash");  // bug — should be gemini-3.5-flash
}

The AI Studio (USE_GEMINI) path correctly uses gemini-3.5-flash, but the Vertex AI path uses gemini-3-flash for both slots.

Workaround

Patch the bundle at chunk-LSXUKR6W.js line 375645, changing:

setFlashModels("gemini-3-flash", "gemini-3-flash");

to:

setFlashModels("gemini-3.5-flash", "gemini-3.5-flash");

Environment

• Gemini CLI version: 0.45.2
• Auth: Vertex AI
• OS: macOS (Darwin 25.4.0, arm64)