Today at the 2024 vivo Developer Conference, vivo officially unveiled its new BlueCore Matrix during the opening keynote. This matrix has been comprehensively upgraded with enhanced capabilities for language models and edge-side models, introducing vivo's proprietary voice model, image model, and multimodal model.
It is understood that the new BlueCore Matrix includes language models, edge-side models, voice models, image models, and multimodal models.
Among these, vivo has introduced a new 3 billion parameter BlueCore edge-side model 3B, which the company claims can rival industry 7B-9B models in capabilities such as dialogue writing, summary generation, and information extraction. Compared to the BlueCore 7B, the BlueCore 3B sees a 300% performance boost, a 46% reduction in power consumption in balanced mode, a 63% reduction in memory usage, and achieves an extreme output speed of 80 characters per second.
Additionally, vivo has launched its new proprietary BlueCore voice model, which can accurately understand natural language, mimic human voices, convey emotions, and support simultaneous interpretation. The BlueCore image and multimodal models have also been upgraded, with the BlueCore image model enhancing the integration of Chinese characteristics and Eastern aesthetics, supporting traditional Chinese ink painting styles, and allowing for the addition of Chinese characters during image creation. The BlueCore voice model, image model, and multimodal model have all ranked first on multiple professional benchmarks.