China's authoritative large model evaluation benchmark SuperCLUE has released its latest comprehensive report. The evaluation results show that ByteDance's Doubao-pro performed strongly, successfully entering the first tier of global large models and competing directly with international top models.

Additionally, Xiaomi's secretly developed MiMo large model also appeared on the list for the first time, drawing industry attention to the capabilities of smartphone manufacturers' self-developed large models.

image.png

Chinese Models Evolve Together: Multi-Dimensional Capabilities Match GPT-4

In this evaluation, domestic large models have made significant progress in understanding Chinese context, common sense reasoning, and logical deduction. Doubao not only performed excellently in basic dialogue quality but also received high scores for its stability in complex task planning and long-text processing.

Baidu ERNIE Bot, Alibaba Tongyi Qianwen, and other models remained at the forefront, showing the deep accumulation of leading companies in corpus accumulation and alignment technology.

Notably, Xiaomi's MiMo appearing on the list indicates that the path of combining edge-side AI with cloud-based large models is becoming viable, offering more possibilities for future smartphone interactions.

Differentiated Competition: From General Intelligence to Vertical Scenarios

The evaluation report points out that the current competition among Chinese large models is no longer just about parameter volume, but rather a more refined scenario-based competition.

Doubao, leveraging ByteDance's ecosystem traffic advantage, performed outstandingly in content creation and social interaction scenarios; while Xiaomi MiMo demonstrated unique advantages in system-level scheduling and multi-device collaboration.

SuperCLUE experts believe that as model capabilities become more balanced, the key to future success will depend on who can more effectively solve industry-specific pain points and provide lower-latency, more cost-effective computing power services.