← Ash FoundryAPI Usage Notes
GitHub Repo

Diagnostics Artifact

Gemini API Tester

A browser-side tester for Google AI Studio / Gemini API free-tier candidates. Paste your own API key locally, select models, and ping them directly without storing the key in the repo.

Manual API key entry Client-side fetch only April 19, 2026

What this page is testing

This artifact is aimed at the Google AI Studio / Gemini API free tier, especially in the context of the Google Cloud trial expiring. The point is not to guess what should work. The point is to let you test real model responsiveness directly from the browser.

The page uses the Gemini REST endpoint and sends a tiny prompt to each selected model. If a model responds, it is reachable with your key. If it fails, the error block should usually tell you whether the issue is invalid auth, model access, quota exhaustion, or a missing free-tier entitlement.

Best-current candidate model IDs

Based on current Google documentation, recent public references, and our own recent local testing, these are the main API model IDs worth probing individually:

  • gemini-2.5-flash
  • gemini-2.5-flash-lite
  • gemini-2.5-pro
  • gemini-2.5-flash-image (image route, worth testing directly even if quota-blocked)
  • gemini-2.5-flash-preview-image (the backend image model name surfaced by quota errors)
  • gemini-3.1-flash-lite-preview
  • gemini-3.1-flash-image (publicly visible in some Google surfaces, worth probing empirically)

Important caveat: some model names are stable public IDs while others are preview or backend-facing names that may 404, 403, or appear only in quota diagnostics. This tester is intentionally empirical, not trusting docs alone.

Run the test

How the tester works

  • Your API key stays in the browser tab. This page does not submit it to the repo.
  • Each selected model is called at v1beta/models/{model}:generateContent.
  • Text models get a tiny text-only ping. Image-labeled models get a multimodal request with responseModalities set to text and image so the route itself can be tested honestly.
  • The result area reports HTTP status, raw error text when available, and any returned text or image-signal metadata.

Interpretation guide

  • 200 + short text response: the text model is reachable with your key.
  • 200 + image part present: the image route is genuinely alive.
  • 400/404: the model ID may be wrong or not exposed on your endpoint/tier.
  • 403: the key is valid but the model is not available to your project or tier.
  • 429: quota or rate limit wall. This is the current expected failure mode for our Google image route.
  • Timeout / fetch failed: network issue, CORS issue, or unstable upstream behavior.