“The same large language model, referred to as 'OPUS 4.6', can produce significantly different outputs for the exact same prompt when run through different platforms like Cursor versus Cloud Code.”