Microsoft has confirmed that OpenAI has been utilizing Looped Transformers in their GPT-6 series models. This confirmation, initially found on a Microsoft webpage, validates previous reporting on the matter. The GPT-6.1 Sol model reportedly uses two inference passes, a reduction from previous iterations, and is built upon the same base model weights as GPT-6 Sol, with differences arising from post-training and the reduced number of loops. However, Microsoft has since removed this information from their web page. AI
IMPACT Confirms a specific architectural choice in OpenAI's GPT-6 series, potentially influencing future model development and understanding of transformer efficiency.
RANK_REASON The cluster reports on a confirmation of a technical detail about a model series, but the primary source (Microsoft webpage) was updated to remove the information, making it commentary on a retracted statement rather than a direct release.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
