Rumors Suggest OpenAI Has Pre-trained 10-Trillion-Parameter Model Codenamed 'Bel'
According to online rumors, OpenAI has completed the pre-training phase for a next-generation AI model codenamed "Bel." The model is reportedly the successor to "Doug" and features over 10 trillion parameters. If true, "Bel" could serve as the foundation for post-GPT-6 models and push the boundaries toward Artificial General Intelligence (AGI). This massive scale could also pressure competitors like Anthropic, who are reportedly facing compute limitations. The rumor originated from an X user, and OpenAI has not officially confirmed the model, released a system card, or announced API access. The model's 10-trillion-parameter size is comparable to rumored sizes for GPT-4.5 and will require further reinforcement learning (RL) to become a finished product.
## BACKGROUND
Pre-training is the initial phase of training a large language model on massive datasets to learn language patterns, which is later refined using reinforcement learning. Parameters are the internal variables that define a model's capacity, with larger parameter counts generally enabling more complex reasoning and knowledge retention.