These benchmark results are currently self-reported from a single source, the evaluation sets are limited in size (30 HumanEval questions, a 502-question MMLU subset), and they have not yet been independently reproduced and cross-validated by the community.