Physix FrontierPhysix Frontier · News Briefing Card (IT Home · Oct 5, 2026)
Bloomberg: China-US top model LiveBench gap narrows to 3%
KEY FACTS
- A Bloomberg Intelligence report dated October 5 says the performance gap between top Chinese and US AI models has fallen to a historic low.
- The report notes that after DeepSeek released V4.1 Flash in September, China's top model trails the US by just 3% in benchmark scores.
- That gap has narrowed sharply from about 9% in May and 15% earlier this year.
- DeepSeek V4.1 Flash ranks sixth globally on LiveBench with a score of 81.1.
- Anthropic holds the highest score at 83.4, and only 3 Chinese models appear in LiveBench's top 15.
KEY DATA
3%China-US top model LiveBench score gap
about 9%May score gap
81.1 pointsDeepSeek V4.1 Flash score
83.4 pointsAnthropic highest score
PHYSIX OBSERVATION
A 3% gap is a signal, not a conclusion. Chinese teams have closed the score gap through optimization on domestic hardware, showing that export controls failed to lock down model capability. But with only 3 Chinese models in the top 15, and leaderboard rank being a different matter from commercial monetization, the pursuers still need to prove they can stay at the table.
Source: IT Home report