Ring-1T

About

Ring-1T is a trillion-parameter mixture-of-experts reasoning model from inclusionAI, released with open weights under the MIT license. The model card describes a thinking mode aimed at deep reasoning and credits two pieces of in-house training infrastructure: Icepop, for keeping reinforcement learning stable over long training runs, and ASystem, the RL framework used at trillion-parameter scale. It runs through Transformers with remote code, and can be served with vLLM or SGLang behind an OpenAI-compatible API.