The latest SuperCLUE benchmark, released on March 30, evaluated 22 AI models across six core tasks, showing that Chinese AI models are closing the gap with Western counterparts.
ByteDance’s Doubao-Seed-2.0-pro achieved an overall score of 71.53 points, just 0.95 points behind OpenAI’s GPT-5.4. The assessment highlights China’s growing competency in AI, particularly in mathematical reasoning and multimodal capabilities.
The global top three models were Anthropic’s Claude-Opus-4.6, Google’s Gemini-3.1-Pro, and GPT-5.4. Doubao-Seed-2.0-pro ranked in the global top five for intelligent agent task planning, surpassing several Western models.
Since its February 2026 release, Doubao-Seed-2.0-pro has attracted interest for its competitive pricing while matching leading Western models in performance. ByteDance’s Doubao app now boasts over 155 million weekly active users.
Xiaomi’s MiMo-V2-Pro scored 60.67 points overall and excelled in mathematical reasoning with a score of 84.03, ranking high among closed-source models. MiMo-V2-Pro, launched on March 18, employs a mixture-of-experts architecture with more than one trillion parameters and supports context windows of up to one million tokens.
The model’s pricing is significantly lower than that of competitors, charging $1 per million input tokens and $3 per million output tokens. MiMo-V2-Pro ranks eighth globally on the Artificial Analysis Intelligence Index and second among Chinese models.
The SuperCLUE evaluation also showcased the strength of Chinese open-source models, with Kimi-K2.5-Thinking and Qwen3.5-397B taking the top three spots in the open-source category. Xiaomi’s open-source model MiMo-V2-Flash, released in December 2025, ranked eighth globally in coding benchmarks by BenchLM.
The March 2026 results reflect a strategic shift among Chinese AI labs from “parameter competition” toward enhancing practical capabilities in reasoning and code generation, indicating a rapid convergence to global standards.
