What changed on the leaderboard
The U.S. edge in AI performance has shrunk to its narrowest level yet, according to a Monday report from Bloomberg Intelligence. After DeepSeek rolled out V4.1 Flash in September, top Chinese models sit roughly 3% behind U.S. peers on benchmark results, a step in from around 9% in May and 15% earlier this year. Lea says the momentum points to more share gains for Chinese players and raises doubts about how effective U.S. export limits on gear like Nvidia Corp. chips will be at slowing China's progress, including efforts by Huawei Technologies Co. to build its own alternatives.
LiveBench, which evaluates models by how they reason through questions, puzzles and tasks (think a proxy for human IQ), put DeepSeek's V4.1 Flash at 81.1 last month, good for sixth globally. As a result, no Chinese system has placed higher since the startup's R1 reasoning model in 2025. Anthropic's best result is 83.4. Lea says the gap now amounts to "comparable performance" with leaders from Anthropic and OpenAI, though only three of LiveBench's top 15 models were Chinese.
Why the gap is closing
Lea ties China's climb to growing technical depth and the ability of local teams to squeeze more from domestic hardware. Lower-cost Chinese systems are also catching up on user adoption just as OpenAI and Anthropic tout superior capabilities while aiming for stock market debuts at trillion-dollar valuations. The bigger picture, in Lea's view, is that the recent gains call into question the durability of U.S. dominance over time.
A closing performance gap changes who captures the economics of the AI build-out. Market Briefs follows that race free every weekday.
Commercial and regulatory headwinds for Chinese firms
Leaderboards are fickle, and scores alone do not translate to revenue. Lea projects China's AI industry may not turn profitable until 2030. A heavy tilt toward low-margin token supply and a bruising price war could prevent any single company from breaking out in a home market where the tally of large language models tops 1,100.
ByteDance Ltd.'s Doubao leads in monetizing AI apps for now, whereas the chatbots offered by DeepSeek and Tencent Holdings Ltd. are still available at no charge. On top of that, Chinese models face growing U.S. regulatory scrutiny and possible bans tied to allegations of model distillation. "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," Lea said.
What it means for your money
Capability is converging, prices are lower on the Chinese side, and usage is edging closer too. But profitability looks a long way off for many China-based players, and the policy risk is getting heavier. Net result for your watchlist: faster catch-up on performance, tighter pricing, uneven monetization, and a geopolitical wildcard that keeps moving.
If the models converge, the advantage shifts to whoever controls cost and distribution. Join Market Briefs free and watch the gap.
