According to CoinMeta, OneMillion.AI reported that less than a week after the release of Jev, the open-source community had already begun to replicate this model, which “does not provide answers but directly gives probabilities.” There are four main approaches: 1. Some directly use the existing qwen model to read the scores for each option; 2. Others simple and jev further eliminate redundant calculations by sharing the same material across multiple questions; 3. Some laya employ traditional classifiers, which result in lower accuracy; 4. von use specialized decision-making models with about 400 million parameters and do not generate text; 5. verdict adopt a smaller approach to handle confidence issues in incorrect answers; 6. kev retain the large model, achieving an accuracy rate of around 78%; 7. nimble focus on training and achieve an accuracy rate of 90.1%.