The average benchmark gap between US and Chinese flagship models has shrunk from 15% at the start of 2026 to 9% in May and 3% after the release of DeepSeek V4.1 Flash in September, according to estimates from Bloomberg Intelligence. Analysts attribute the gains to Chinese labs optimizing algorithms for domestic accelerators, and on that basis question how much US export controls on Nvidia chips are still limiting. Parity in capability has not translated into parity in revenue: the same forecast expects the sector to stay unprofitable until 2030 because of the ongoing price war, with only a handful of players such as ByteDance earning, while DeepSeek distributes its models for free and sells tokens at near-zero margin.

Bloomberg: US Lead in AI Over China Narrows After DeepSeek Gains, BI Says

Related: DeepSeek Releases V4.1 Flash, Kimi K3 Could Be China’s DeepSeek Moment, DeepSeek Developing Its Own AI Chip