According to Bloomberg Intelligence, the performance gap between top Chinese AI models and their American counterparts has narrowed to 3% as of October 5. This follows the release of the DeepSeek V4.1 Flash model in September, which significantly improved the competitive standing of Chinese AI labs.
The performance difference has decreased consistently throughout the year, dropping from 15% earlier in 2024 to 9% in May, and finally to 3% in September. Analyst Robert Lea notes that these advancements challenge the long-term technological dominance of the United States in the artificial intelligence sector.
DeepSeek V4.1 Flash secured the sixth position in the global LiveBench rankings last month. The model achieved a score of 81.1 points, trailing closely behind the top-performing Anthropic system, which scored 83.4 points. This marks the highest ranking for a Chinese model since the R1 reasoning model breakthrough in 2025.
The rise of Chinese AI capabilities is attributed to deeper technical expertise and the optimization of models for domestic hardware. Despite the narrowing performance gap, Chinese models account for only three of the top 15 positions on the LiveBench leaderboard, and analysts warn that monetization remains a challenge.
