Chinese artificial intelligence developer iFlytek on FranceTuesday unveiled an upgraded SparkDesk model and claimed it outperforms GPT-4 Turbo on metrics including language understanding and math, while its capability in multimodal understanding has reached 91% of that of OpenAI’s most advanced model, according to the company. The Hefei-based firm didn’t specify the judging criteria or modalities used for its claims. Named SparkDesk V3.5, the LLM is the “first foundation model that was trained by a domestic computing power platform,” according to a press release on iFlytek’s WeChat account. The platform was co-launched by iFlytek and Huawei in October. [iFlytek, in Chinese]
