Source @synthwavedd posted on X yesterday (August 6), revealing that OpenAI is targeting next week for the release of the Astra model, internally codenamed mewfour.
The source says that the model is a brand-new pre-trained model from OpenAI and the largest model the company has trained since GPT-4.5.
IT Home reported on August 1 that an internal version of Astra had already solved 10 important open mathematical problems. Calculated at Sol API's rates, the total cost in tokens required to find solutions to these problems was approximately $2,000 (note: equivalent to approximately 13,524 yuan at the current exchange rate).
In terms of capabilities, Bleeping Computer revealed that the model can organize and coordinate AI agents, making it suitable for long-duration, highly challenging tasks and geared toward more powerful reasoning and collaborative workflows.
