AI Agents - Google Gemini 2.0 Flash is pointing to a nonexistent model

[dominikmucklow]

Can we access your project?

• I give permission for members of the FlutterFlow team to access and test my project for the sole purpose of investigating this issue.

Current Behavior

When using an AI Agents set to Google Gemini 2.0 Flash this error appears:

Error: Publisher Model 'projects.../models/gemini-2.9-flash-lite-preview-02-5' not found.

This does not occur when a different model is selected, such as Gemini 2.0 Flash Thinking.

Expected Behavior

I think the correct model name is "gemini-2.0-flash" or "gemini-2.0-flash-lite".

Steps to Reproduce

Create an agent with Gemini 2.0 Flash, and run it in test mode to get the error.

Reproducible from Blank

• The steps to reproduce above start from a blank project.

Bug Report Code (Required)

gemini-2.0-flash-lite-preview-02-05:generateContent:1 Failed to load resource: the server responded with a status of 404 ()

Visual documentation

Environment

- FlutterFlow version: v5.3.5
- Platform: Web
- Browser name and version: Chrome
- Operating system and version affected: macOS Sonoma 14.1

Additional Information

No response

[Alezanello]

Thank you so much for flagging this issue! I’ll go ahead and share it with our engineering team for further review.

[MaggieThomann]

Yes, thanks for reporting! This will be fixed in today's release.

[FredvanRijswijk]

issue is coming up again, also I noticed it's not using the region from my setup europe-west3 but always us-central1

[MaggieThomann]

@FredvanRijswijk

This issue was fixed but then Google updated their model names again. I have a fix that should go live tomorrow that will support the stable models Google currently offers via Firebase Vertex AI (Gemini 1.5 Pro & Flash and Gemini 2.0 Flash and Flash Lite) in addition to the experimental models (Gemini 2.5 Pro & Gemini 2.0 Flash Thinking).

In the future, we'll also be able to make these model name updates outside of our typical release cadence so that we can fix these issues much faster.

For now, after tomorrow, I would recommend using Google's stable models as listed here

[FrancoSernaglia]

Hi, is the solution ONLINE yet?

[MaggieThomann]

Apologies, we postponed our deploy by 1 day to today @FrancoSernaglia, so it should be available today once we deploy. I'll close the issue once we deploy!