SuperCLUE November benchmark puts Doubao first in China at 71.53 points
In the November SuperCLUE results covering 22 models, Doubao-Seed-2.0-pro scored 71.53 to take first place in China, 0.95 points behind GPT-5.4.
Chinese large model benchmark SuperCLUE published its latest November 2026 results, with 22 mainstream domestic and international models competing and data as of November 2026.
ByteDance Doubao-Seed-2.0-pro-260215 took first place in China with 71.53 points, entering the global first tier. The benchmark covers six core tasks including mathematical reasoning, scientific reasoning and code generation. Overseas closed models still led the overall ranking, with Anthropic Claude Opus 4.6 max, Google Gemini 3.1 Pro Preview high and OpenAI GPT-5.4 xhigh occupying the top three places. Doubao followed closely, trailing GPT-5.4 by just 0.95 points, and overtook some overseas models in agent task planning to reach the global top five.
Two SenseNova models from SenseTime made the board, with SenseNova 6.7 Flash-Lite scoring 60.67 among leading closed models and 84.03 on mathematical reasoning, while SenseNova U1 Fast strengths lay in multimodal understanding and visual content generation. SenseTime positions the product on long-chain enterprise office tasks, and by its published figures the natively multimodal architecture places visual capability in the model base, cutting token consumption by about 60 percent in information search scenarios versus text-only agents.
The results also show domestic open-weight models taking the top three open-source places, with Kimi-K2.5-Thinking and Qwen3.5-397B-A17B-Thinking well ahead of comparable overseas models. The benchmark organizers noted that the overall ranking reflects general task performance, that model positioning differs across vendors, and that single-task scores fluctuate with version iteration, so it should be read against specific business scenarios.