# Reasoning Engine (Preview) queries fail with 404 Publisher Model for Gemini models (us-central1, europe-west1)

Google AI Studio
,
1

Hi all — looking for guidance from the community and Google folks on a blocking issue with Vertex AI Reasoning Engine (Preview).

## Summary

- Deployments succeed, but all queries fail with:

```

Reasoning Engine Execution failed.

Error Details: {“detail”:“Agent Engine Error: An error occurred during invocation. Exception: 404 Publisher Model `projects/PROJECT_ID/locations/{region}/publishers/google/models/gemini-1.5-flash[-002]` was not found or your project does not have access to it.”}

```

- Reproducible in both `us-central1` and `europe-west1`

- Affects multiple engines and both `gemini-1.5-flash` and `gemini-1.5-pro` (versioned and unversioned)

## Environment

- Project: `PROJECT_ID` (number: `PROJECT_NUMBER`)

- Regions tried: `us-central1`, `europe-west1`

- SDK: `google-cloud-aiplatform==1.83.0`

- Python: 3.10 (venv), 3.12 (system)

- Staging bucket: `gs://STAGING_BUCKET_NAME`

## Minimal Repro

```python

import vertexai

from vertexai.preview import reasoning_engines

from vertexai.generative_models import GenerativeModel

class BriefAgent:

def _init_(self):

# Hardcode init in case cloudpickle loses state in runtime

    vertexai.init(*project*="PROJECT_ID", *location*="us-central1")

self.model = GenerativeModel(“gemini-1.5-flash”) # also tried “gemini-1.5-flash-002”, pro variants

def query(self, prompt: str):

    system_instruction = "You are a helpful research assistant..."

    resp = self.model.generate_content(\[system_instruction, prompt\])

return resp.text

# Deploy (succeeds)

vertexai.init(project=“PROJECT_ID”, location=“us-central1”, staging_bucket=“gs://STAGING_BUCKET_NAME”)

engine = reasoning_engines.ReasoningEngine.create(

BriefAgent(),

display_name=“zen-l1-v002”,

requirements=[“google-cloud-aiplatform==1.83.0”, “cloudpickle==3.1.1”],

extra_packages=[“l1_as_reasoning_engine_inline.py”],

)

print(engine.resource_name)

# Query (fails with 404 Publisher Model)

remote = reasoning_engines.ReasoningEngine(engine.resource_name)

print(remote.query(prompt=“Key trends in quantum computing?”))

```

## Affected Engines (examples)

- us-central1:

- `projects/PROJECT_NUMBER/locations/us-central1/reasoningEngines/ENGINE_ID_1` (zen-l1-v002)

- `projects/PROJECT_NUMBER/locations/us-central1/reasoningEngines/ENGINE_ID_2` (zen-l1-intuitive-brief)

- `projects/PROJECT_NUMBER/locations/us-central1/reasoningEngines/ENGINE_ID_3` (zen-l1-us)

- europe-west1:

- `projects/PROJECT_NUMBER/locations/europe-west1/reasoningEngines/ENGINE_ID_4` (zen-l1-working)

- and several others with the same behavior

## What we tried

- Model names: `gemini-1.5-flash`, `gemini-1.5-flash-002`, `gemini-1.5-pro`, `gemini-1.5-pro-002`

- Regions: `us-central1` and `europe-west1`

- Ensured `vertexai.init()` is called both at deploy-time and inside agent `_init_`

- Removed unsupported `location` argument from `GenerativeModel()`

- APIs enabled: `aiplatform.googleapis.com`, `generativelanguage.googleapis.com`

- IAM: granted to runtime service accounts (attempted)

- `service-PROJECT_NUMBER@gcp-sa-aiplatform-re.iam.gserviceaccount.com`: `roles/aiplatform.user`, `roles/serviceusage.serviceUsageConsumer`

- `service-PROJECT_NUMBER@gcp-sa-aiplatform.iam.gserviceaccount.com`: same

- Local validation: same code works outside Reasoning Engine (e.g., Cloud Run), only fails in Reasoning Engine runtime

## Questions

  1. Do Reasoning Engine runtimes require additional roles or special enablement to access Gemini publisher models?

  2. Are there region or model restrictions for Reasoning Engine (Preview) we should be aware of?

  3. Should we reference models by full resource path instead of short name inside Reasoning Engine?

  4. Is this a known Preview limitation? Any recommended workaround beyond Cloud Run?

## Impact

- Blocks moving L1/L4 agents to Reasoning Engine Preview; Cloud Run works but loses managed sessions/memory.

## Related docs

- Troubleshooting (Deploy): https://cloud.google.com/vertex-ai/generative-ai/docs/agent-engine/troubleshooting/deploy

- Troubleshooting (Use): Troubleshoot using an agent with Vertex AI Agent Engine Runtime  |  Vertex AI Agent Builder  |  Google Cloud

- Model versions/regions: Model versions and lifecycle  |  Generative AI on Vertex AI  |  Google Cloud

- Forum category: Google AI studio

Thanks in advance! If helpful, I can provide more engine IDs or logs.