OpenAI内部代码泄露，最强模型「皇帝」登基，0思考延时吓人

Core Insights - OpenAI is preparing to launch four new models, collectively referred to as the "Penguin family," with varying inference budgets, indicating a strategic enhancement in their AI capabilities [1][10] - The new models include Emperor (512), Rockhopper (64), Macaroni (16), and Mumble (0), with Emperor being the flagship model [3][10] - A new "memory search" feature is being developed for ChatGPT, aimed at improving user experience by allowing quick retrieval of stored memory information during conversations [5][21] Model Developments - The Emperor model has the highest inference budget of 512, positioning it as a state-of-the-art flagship model [3][10] - The Rockhopper model, with an inference budget of 64, serves as a mid-tier option, while Macaroni is designed for quick and efficient inference with a budget of 16 [10] - Mumble, with no inference budget, is designed for rapid responses, showcasing a diverse range of capabilities within the Penguin family [10] Memory Search Feature - The "memory search" function aims to streamline the user experience by allowing users to retrieve relevant memory entries through direct prompts, addressing current usability issues [21][22] - This feature is expected to enhance workflow efficiency and context retrieval, similar to functionalities found in other applications like the Atlas browser [22][24] - The introduction of this feature is seen as a strategic move to retain existing users and attract new ones amid increasing competition in the AI assistant market [24][26] Competitive Landscape - OpenAI's market share has been impacted by competitors like Google's Gemini 3, which has led to a 6% decline in daily active users for ChatGPT [24] - There are speculations that OpenAI may expedite the release of GPT-5.2 in response to competitive pressures, potentially launching it sooner than expected [29] - The company is also facing backlash from users regarding the introduction of ads in ChatGPT, which has led to subscription cancellations [26][28]