Yuanxiang Technology announced that it will open source its high-performance large model XVERSE-65B-2 base version, which is a landmark event. The model features specialized improvements in coding and math capabilities while retaining strong capabilities for generative creation, role-playing, and tool usage. It is worth noting that XVERSE-65B-2 has achieved excellent results in various comprehensive evaluations, surpassing many existing models and becoming the leader among domestic open source models. This not only reflects Yuanxiang Technology’s strong technical strength in the field of AI, but also makes an important contribution to promoting the development and application of large model technology. More importantly, as the earliest free commercial model in China that is open source and has the largest parameters, XVERSE-65B-2 will provide convenience to more developers and enterprises.
Yuanxiang announced the open source high-performance large model XVERSE-65B-2 base version, specifically improving its coding and mathematical capabilities, and maintaining the powerful capabilities of generative creation, role-playing and tool use of the previous generation. This model surpassed GPT3.5 and Llama2-70B in comprehensive evaluation, and comprehensively surpassed open source benchmarks. In the latest comprehensive evaluation of the SuperCLUE Chinese universal large model, XVERSE-65B performed outstandingly among domestic open source models. Yuanxiang has open sourced three large models, among which XVERSE-65B is the earliest open source and free commercial model with the largest parameters in China.
The open source of XVERSE-65B-2 will further promote the development of domestic large model technology, lower the development threshold, provide convenience for more developers, and promote the popularization and application of AI technology. It is believed that Yuanxiang Technology will continue to launch more excellent large-model products in the future and contribute more to the field of artificial intelligence.