The apparent bias may be explainable as it’s not remarkable for OpenAI or Anthropic to be slightly ahead. It _is_ remarkable for an open weights model to be better than the closed models from the trillion dollar (allegedly) companies.