Whoa, according to the latest leaks, OpenAI's new model codenamed Astra is set to launch next week!!! The internal checkpoint recently used for Dogfood testing is named 'mewfour,' and it has now been selected as the Release Candidate. Astra is not a minor iteration like GPT-5.6. It’s a completely new pre-trained model and represents the largest-scale base model training by OpenAI since GPT-4.5. In recent times, advancements in cutting-edge models have mostly come from post-training, inference-time computation, tool integration, and Agent system optimization. Astra, however, might signal OpenAI raising the fundamental capability ceiling of its models again through larger-scale data and compute power. In other words: GPT-5.6 is more about organizing existing intelligence more efficiently, while Astra could be about creating an entirely stronger brain. The key leaked info so far boils down to three points: - A brand-new pre-trained model, not just fine-tuning an old one - The largest scale since GPT-4.5 - 'mewfour' checkpoint has been chosen as the release candidate