iFlytek Unveils Spark Cognitive Model V3.5: Surpasses GPT-4 Turbo in Multiple Abilities
At the Xunfei Spark Cognitive Model V3.5 upgrade release conference on January 30th, iFlytek officially announced the latest version of the Xunfei Spark Cognitive Model V3.5.
According to iFlytek, the Xunfei Spark Cognitive Model V3.5 has achieved significant improvements in multiple aspects. Specifically, the text generation capability has increased by 7.3%, the language understanding capability has increased by 7.6%, the knowledge question-answering capability has increased by 4.7%, the logical reasoning capability has increased by 9.5%, the mathematical capability has increased by 9.8%, the coding capability has increased by 8.0%, and the multimodal understanding capability has also increased by 6.6%.
It is worth mentioning that the Xunfei Spark Cognitive Model V3.5 has surpassed GPT-4 Turbo in language understanding, mathematical capability, and coding capability. Specifically, the mathematical capability of this model is claimed to have surpassed GPT-4 Turbo, and the coding capability has reached 96% of GPT-4 Turbo, while the multimodal understanding capability has reached 91% of GPT-4V.
To support the powerful functionality of the Xunfei Spark Cognitive Model V3.5, iFlytek and Huawei jointly developed the domestic computing platform "Feixing One," which was officially launched on October 24, 2023. It is said that this platform enables Huawei's Ascend 910B to achieve 90% of the training efficiency of NVIDIA A100.
Looking ahead, iFlytek stated that they plan to elevate the Xunfei Spark Cognitive Model to the current best level of GPT in the first half of 2024 and achieve surpassing performance in industry applications.