A user on Reddit has shared a dataset of 25 billion tokens, reportedly used for training GPT-5.5. The dataset includes over 12,000 tool calls, with a link provided to a Hugging Face repository for the data. This release appears to be a community-driven effort to compile training data for a hypothetical or upcoming model. AI
RANK_REASON The item is a user-generated dataset shared on Reddit, not an official release or announcement from a frontier lab.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →