DeepSeek V4.1 Flash narrows US-China AI gap to 3% on LiveBench

DeepSeek's V4.1 Flash, released September 10, scored 81.1 on LiveBench against 83.4 for Anthropic's leading model, a 2.3-point spread Bloomberg Intelligence calls the smallest US-China top-model gap it has tracked.

The roughly 3% differential compares with about 9% in May and 15% earlier in 2026, according to a note Monday from BI senior analyst Robert Lea. On the agentic coding sub-benchmark, DeepSeek's 77.3 actually ran ahead of Anthropic's 66.1, per the October 4 leaderboard snapshot reported by Implicator.

Lea attributes the catch-up to Chinese labs improving technical capabilities and optimizing models for domestic hardware, which he says raises questions about the effectiveness of US chip export controls. Only three of the top 15 LiveBench entries are Chinese, so the gains sit at the frontier rather than across the field.

The commercial picture is less rosy. China's market now hosts more than 1,100 large language models, and Lea does not expect the sector to turn profitable before 2030. "Putting China's AI sector on a sustainable profit footing will require a cooling of competitive pressures, an industry shakeout, and a more rational approach to pricing," he wrote.

DeepSeek has been one of our busier trackers this autumn, between coverage of resold Claude access in Beijing storefronts and Reflection AI's forthcoming open-weight challenger, both pointing at the same thinning Western moat.