OpenAI Tests 'Emperor' Model, 'Penguin' Family & ChatGPT Mem

OpenAI is reportedly testing a new suite of models, internally referred to as the "Penguin family," alongside a "memory search" feature for ChatGPT. These developments suggest an accelerated product roadmap for the company.

Internal code screenshots indicate four models within the Penguin family, each with varying inference budgets: Emperor (512), Rockhopper (64), Macaroni (16), and Mumble (0). The "Emperor" model, with the highest inference budget, is speculated to be a new flagship offering. Inference budget, or "Juice," refers to the computational resources available to the model for a single conversation.

New Models and Features Under Development

The "Penguin family" models are reportedly undergoing testing on DesignArena. The "Emperor" model, with its substantial inference budget, is anticipated to be a high-performance model. "Rockhopper" is positioned as a mid-range model, while "Macaroni" is designed for efficiency. "Mumble," with zero inference, is expected to offer extremely fast response times. The release timeline for these models remains undisclosed.

Separately, OpenAI is preparing a "memory search" mode for ChatGPT. This feature aims to improve context retrieval by allowing users to directly query stored conversational memory. The current memory management interface is considered difficult to navigate, particularly with numerous entries. The new search function would enable ChatGPT to retrieve relevant memories based on user prompts, streamlining workflow and context management. This functionality bears resemblance to the "Browser Memory" feature found in the Atlas browser. While the memory search function appeared briefly and is not yet publicly available, its presence suggests internal testing or preparation for a phased rollout.

Previous reports from The Information indicated that OpenAI plans to launch a new inference model soon, potentially codenamed "Shallotpeat." This model is said to have outperformed Google's Gemini 3 in internal evaluations. Another model, "Garlic," has reportedly completed pre-training and is expected to be released as GPT-5.2 or GPT-5.5 early next year.

These developments occur as OpenAI faces increased competition. Following the release of Google's Gemini 3, OpenAI reportedly saw a 6% drop in ChatGPT's daily active users. Additionally, some paid ChatGPT users have reported encountering advertisements during conversations, leading to cancellations of subscriptions. Some observers speculate that OpenAI may accelerate the release of GPT-5.2 in response to these competitive pressures and user feedback.