Bloomberg Intelligence says top Chinese AI models now trail US rivals by about 3 percent on benchmarks
The gap was roughly 9 percent in May and about 15 percent earlier this year, and it narrowed after DeepSeek released V4.1 Flash in September.
Top Chinese AI models now trail their US rivals by about 3 percent on benchmark scores, according to a Bloomberg Intelligence report published on October 4 by senior analyst Robert Lea, as described by Startup Fortune. The gap was roughly 9 percent in May and about 15 percent earlier in the year. The latest narrowing followed DeepSeek's release of its V4.1 Flash model in September. Bloomberg Intelligence says the trend points to further market share gains for Chinese AI labs. Startup Fortune argues that a gap this small makes it harder to justify paying a premium for American frontier models. The report recalls a precedent: in January 2025, the release of DeepSeek's R1 model contributed to a record sell-off in Nvidia shares. On January 27, 2025, Nvidia lost $589 billion in market value, the largest one-day loss in US stock-market history, although the stock later recovered. The new figures matter for US companies that have spent heavily to stay ahead of Chinese competitors, and for buyers comparing the price and performance of US and Chinese models. The report summary does not say which benchmarks were used to measure the gap.