DeepSeek V4.1 Flash narrows AI gap with US to 3 percent

The performance gap between top Chinese artificial intelligence models and their American rivals has narrowed to approximately 3 percent, according to recent benchmark data. This convergence follows the September launch of DeepSeek V4.1 Flash, a significant improvement from the 9 percent deficit recorded in May and the 15 percent spread observed at the start of the year.

Shrinking technical divide

Bloomberg Intelligence senior analyst Robert Lea stated that this accelerated progress could enable Chinese firms to capture a larger share of the global AI market in the coming years. The narrowing disparity is attributed to deepening local expertise and the ability of Chinese researchers to optimize models for domestic hardware efficiency.

This development raises questions about the effectiveness of US export controls on advanced technologies, including Nvidia chips. These restrictions were originally designed to slow China’s AI advancement and prevent companies like Huawei from developing competitive local alternatives.

Global influence contest

Lea noted that China’s progress "raises further doubts about the sustainability of US technological superiority in artificial intelligence over the long term." The two nations are competing for dominance in a sector deemed critical for economic productivity, military advantage, and geopolitical influence.

While US companies such as OpenAI and Anthropic seek to justify market valuations potentially reaching trillions of dollars by highlighting model superiority, lower-cost Chinese systems are rapidly closing the performance and user-base gaps.

LiveBench rankings shift

DeepSeek V4.1 Flash climbed to sixth place globally on the LiveBench index last month. This marks the highest ranking for a Chinese model since the R1 reasoning model achieved success in 2025. The LiveBench metric evaluates AI capabilities in analyzing questions, solving problems, and executing complex tasks, functioning similarly to human cognitive assessment.

The DeepSeek model scored 81.1 points, compared to 83.4 points for Anthropic’s leading systems, indicating near-parity with top-tier global platforms. Despite the reduced 3 percent gap, Chinese entities still account for only three of the world’s top 15 models on the LiveBench leaderboard.

Profitability challenges persist

Although technical advancements are evident, the report suggests the Chinese sector may struggle to convert these gains into profits. Analyst Lea predicts the Chinese AI industry will remain unprofitable until 2030 due to intense price wars, thin margins, and a crowded market containing more than 1,100 large language models.

ByteDance currently leads revenue generation through its Doubao platform, while chatbots from DeepSeek and Tencent remain free for users. Building a sustainable, profitable Chinese AI ecosystem requires easing competitive pressures, according to the analysis.