Personally I use OpenAI models via the API for transcription of PDF files. Is there a big difference between them and Gemini models?