OpenAI is quietly advancing its AI project codenamed "Strawberry," which has made significant technological progress in handling complex problems and logical reasoning. According to insiders, this breakthrough technology is expected to improve the performance of current AI models in solving mathematical problems, programming tasks, and other complex scenarios. The "Strawberry" project aims to lay the foundation for OpenAI's next-generation flagship language model, "Orion," by providing high-quality training data. This strategy is expected to reduce errors and "hallucinations" in AI model responses, thereby enhancing their accuracy and reliability. In addition, OpenAI is also exploring the integration of a streamlined version of "Strawberry" into chat applications to enhance AI's reasoning abilities in deep conversations. According to reports from US media, OpenAI plans to launch the "Strawberry" project this fall, possibly as an upgraded version of ChatGPT. The new AI system will be able to solve various complex tasks, including mathematical problems and word games, which are still challenging for most existing AI systems. The "Strawberry" project not only aims to improve problem-solving abilities but also aims to enhance AI's performance in programming and multi-step task processing. This update is expected to bring more business opportunities to OpenAI. Although the company is currently in a net loss position, its enterprise services and ChatGPT subscriptions have generated a monthly revenue of $283 million. It is worth noting that the "Strawberry" project includes multiple versions, with a simplified version being used for chatbots and a complete version focusing on generating training data for the Orion model to address the scarcity of high-quality internet text. As competitors like Anthropic and Meta catch up, OpenAI needs to further consolidate its leading position in the AI field through the "Strawberry" project. However, compared to the speed-oriented GPT-4, the "Strawberry" model may sacrifice some reaction speed while enhancing reasoning abilities, making it more suitable for scenarios that do not require immediate feedback.