Meta was reportedly concerned that Llama 4 was less capable than OpenAI's models in conducting humanlike voice conversations.