OpenAI Completes Pre-training of Codename ‘Bel’, Potential Basis for GPT-6 and Astra

The completion of pre-training for 'Bel' signals that OpenAI is moving beyond incremental model updates, positioning this as a foundational layer for both GPT-6 and Astra. This matters because it suggests a convergence of capabilities—possibly unifying conversational AI and multimodal or agentic features—which could intensify competitive pressure across the AI stack. For market observers, the key is whether downstream integration into products like Astra accelerates enterprise adoption or raises compute cost concerns. Whether 'Bel' delivers on its foundational promise without major alignment or safety setbacks is the open question.