OpenAI New flagship model Astra will be released as soon as next week: GPT-4

OpenAI will release the new flagship model Astra as soon as next week: the largest pre-training since GPT-4.5, focusing on agent collaboration. Sources broke the news that OpenAI aims to release the new AI model Astra next week, with the internal code name mewfour. It is revealed that Astra is the largest pre-trained model trained by OpenAI after GPT-4.5, marking a major leap in the company's model capabilities. It has been previously reported that an internal version of Astra successfully solved 10 important open mathematics problems. If calculated according to the Sol API rate, the total word cost required to find these solutions is approximately US$2,000, or approximately 13,500 yuan, demonstrating the amazing potential of this model in the field of complex mathematical reasoning. In terms of capability positioning, according to Bleeping Computer, the core highlight of Astra is its ability to organize and collaborate multiple AI agents, which is suitable for long-term and difficult tasks and is geared toward stronger reasoning and collaborative workflows. This design direction shows that OpenAI is expanding its model capabilities from single-round dialogue and generation to multi-agent orchestration and automated execution of complex tasks, which is highly consistent with the current industry trend of transforming from conversational AI to intelligent AI. If Astra does have intelligent agent collaboration capabilities, it is expected to become a key infrastructure for enterprise-level AI applications. The pace of model releases has accelerated, a key step under competitive pressure OpenAI The pace of model releases has accelerated significantly recently, and the timing of Astra's launch is also closely related to the competitive pressure it faces. Google's flagship model Gemini 3.5 Pro has been delayed multiple times, while Anthropic's Claude series continues to encroach on market share in the programming and inference fields. OpenAI needs to consolidate its technology leadership with new models. Choosing to release Astra’s release signal at this time is not only a response to market expectations, but also an important move to convey confidence to developers and enterprise customers. Astra’s breakthroughs in solving mathematical problems are particularly noteworthy. Open mathematics problems have long been regarded as a hard indicator of AI's reasoning ability. Being able to solve 10 such problems means that the model has the ability to in-depth logical reasoning and multi-step deduction, rather than just relying on pattern matching to give approximate answers. Combined with its agent collaboration characteristics, Astra may be positioned as a dedicated model for scientific research, engineering and complex decision-making scenarios, complementing the GPT series of general dialogue models. Can OpenAI use this to reopen the gap with its competitors?