According to CoinMeta, the first team in the ARC competition, Mostik, has recently made its appearance. This team has developed a bridging technology that allows different AI models to directly exchange their internal states, thereby eliminating the need for the generation costs associated with large models. Currently, Mostik is leading in the ARC-AGI-3 Kaggle competition with a score of 7.51%. The team consists of 15 members, including 12 doctors and one Fields Medalist. The second milestone deadline is September 30th, with the final submission date being November 2nd. The technology used by this team involves large models handling pre-filling tasks while smaller models are responsible for decoding, resulting in a computational load that is only about 40% of that required for medium-sized single models with the same accuracy rate. However, the final outcome is still lower than that achieved by the 753B large model itself.