LocalLLaMA leans Qwen for mixed Indonesian text

This is useful practitioner signal, not a definitive benchmark: once multilingual edge cases and code-mixed prompts enter the picture, commenters still trust Qwen more than GPT-OSS.

• –Multiple replies explicitly warned against GPT-OSS for non-major-language work, with one commenter saying its non-English output deteriorates when English and another language are mixed in the same prompt
• –The most nuanced response said GPT-OSS can be good for smaller languages when the data stays monolingual and the task does not need much reasoning, but recommended Qwen or Gemma for mixed-language inputs
• –That distinction matters for real analytics pipelines, where short user-generated rows often contain slang, transliteration, and English loanwords rather than clean benchmark-style text
• –This is best read as field feedback from local-model users, not as a formal model ranking or a new release announcement