I am facing multiple inconsistent and critical issues with my Opal mini-app using the Gemini API (2.5 Pro/Flash models) since this morning (approx 8 AM IST). The app was working perfectly for the last 10 days. The issues are:
Summarization Issue: Hindi to English translation app is now summarizing the text instead of providing a full literal translation.
Wrong Language Output: Hindi to Kannada translation app is returning the final output in English instead of Kannada. This happens even in a two-step workflow (Translate → Generate HTML).
Overload Error: I am frequently getting the error “Model is overloaded” when trying to run the apps.
These issues suggest a potential server load or a recent model behaviour change. Could you please check for any ongoing incidents or maintenance? Thank you for your help!
Thank you for reporting this. To help us debug the issue of inconsistent output, could you please share a minimal reproducible code snippet ?
Regarding the Model Overloaded error, could you also clarify if you are currently using the free or a paid tier? We ask because requests from paid tiers are prioritized especially during high-traffic periods.
Hello Juliana and Chanduriv, thank you both for your responses and ongoing support.
First, an update: I can confirm that the ‘Model is overloaded’ errors and general app loading failures have largely been resolved. The applications are now running without those critical downtime issues.
Tier and Troubleshooting Confirmation:
Tier Status: To clarify my subscription (as Chanduriv asked), I am running the main model on the Gemini Pro plan. However, other Google API services used within the Opal workflow are on the free tier.
Cross-Check: I even created new apps and checked the language issue using a different person’s email ID (on a free tier), and the same translation failures were occurring there too. This suggests the language problem is a widespread issue, not specific to my personal account or free API usage.
The Core Problem Persists (Prompt Adherence Failure):
Despite the stability improvement, the core translation quality issues persist:
Persistent Summarization: The Hindi to English translation is still being heavily summarized, ignoring all instructions for a full, literal translation.
Language Inconsistency: Translation from Hindi to target languages like Kannada or Bengali consistently reverts the final output to English. I am sharing a screenshot below that clearly shows the intended Kannada output failing and producing English summaries instead.
A Personal Note:
I want to sincerely thank the developer team. My app was working beautifully and accurately until yesterday, and I was very happy with the results. However, it is frustrating to face this problem continuously since yesterday (Nov 12th).
Could the team please urgently investigate the Prompt Adherence in the current build of the Gemini model to ensure it respects target language and literal translation instructions?
Thank you again for your time and continued support.
