​I'm running a chatbot that handles NSFW content, and I'm consistently running into issues with the Gemini models' content filters, especially with Gemini 2.5 Pro.

​Here’s the strange part: My service in Korean works relatively fine, with a manageable filter rate. However, my service in Thai triggers the content filter at a massively higher rate—far beyond any normal margin of error.

​I am using the exact same prompt for both languages. The only difference is the language of the user interaction.

​The failure rate difference between the two languages is over 10%, and I can't figure out the cause.

​Has anyone else experienced this? Is there a known issue where the Thai language is significantly more sensitive or triggers Gemini's content filters more frequently than other languages?

​Thanks in advance