Kuaikeji, May 28 - The latest evaluation results for Chinese large language models from SuperCLUE have been released, revealing that the top tier globally is firmly occupied by overseas models, while the top three domestic models are fiercely competing for fifth place worldwide.

This evaluation covered 21 mainstream models from both domestic and international sources, with the evaluation set including six major tasks: mathematical reasoning, scientific reasoning, code generation, agent (task planning), precise instruction following, and hallucination control, totaling 492 questions.

Four overseas models—Gemini, GPT-5.5, Claude-Opus, and Gemini-Flash—securely occupy the top four positions, forming an unshakable top tier.

The scores of the three domestic models—DeepSeek-V4-Pro, Qwen3.7-Max, and Doubao Seed 2.0 Pro—are very close, forming the domestic top tier. Their global rankings are concentrated around fifth place, making them the leading domestic group.

Although there is still an overall gap, the progress of domestic models is evident. In the code generation task, Qwen3.7-Max scored only slightly lower than the top overseas models, with a gap of less than 2 points. In tasks such as mathematical and scientific reasoning, domestic models have also repeatedly ranked among the global leaders.

In terms of cost-effectiveness, domestic models have a distinct advantage, with multiple products entering the high cost-effectiveness range, achieving results close to the top models at a lower cost. However, in terms of reasoning efficiency, overseas models still dominate the high-efficiency range, while domestic models mostly fall into the mid-to-low range, leaving room for improvement.

Overall, domestic large language models are rapidly catching up, but the global top-tier landscape remains unchanged for now.