Xu Li, Chairman and CEO of Shangtang Technology (00020.HK), announced the release of the RiriNova 5.5 version at the WAIC Shangtang Forum, which benchmarks GPT4-Turbo and improves the baseline model performance with 600 billion parameters by 30%. Through synthetic data, it improves reasoning and thinking abilities. After upgrading the client-side model, the model accuracy improved by 10%, the first package delay decreased by 40%, and the inference efficiency increased by 15%.
Xu Li pointed out that through the superimposition of a large amount of industry information and data, the RiriNova 5.5 version's reasoning and language capabilities were improved, and model abilities were improved by an average of 30%. Among them, mathematical reasoning improved by 31.5%, English ability improved by 53.8%, instruction following improved by 26.8%, and the interactive effect exceeded GPT4o in many-core key indicators, surpassing the GPT4o-0513 version.