The thinking model bags from Ant Group are IMO silver

Ring-1T, an open-source "thinking model" developed by Ant Group's AGI program InclusionAI, placed second to Google and OAI's gold-level models on the International Mathematical Olympiad. The specifics: Ring-1T, which is based on Inclusion's MoE design, operates on 1T parameters (50B active) with a 128K context window for multi-turn and ultra-long reasoning. In the IMO test, the model achieved silver-level and almost matched Gemini 2.5 Pro on the hardest question, solving four problems in one attempt and one in three. It outperformed or roughly matched the flagship models from Google, Anthropic, and OpenAI, delivering near SOTA performance across a number of benchmarks. According to InclusionAI, Ring-1T is undergoing active training and development with the aim of enhancing efficiency and alignment as well as enabling deeper thinking. The distinction between open and closed AI is further blurred by Ring-1T, which performs close to SOTA and has reasoning that is nearly on par with GPT-5. It is also a significant step in China's efforts to compete with the West in the AI race, and it will not be long until the nation creates a model that can compete on a global scale.