Run the same task against several explicitly selected open-weight models in parallel and return every answer separately. No ranking, verification, or synthesis.