AIToday
Large Language ModelsAI Regulation & PolicyAI Business & IndustryFortune AIPublished: Oct 8, 2026, 10:01 JST

China's AI models close to four months behind OpenAI and Anthropic

China's AI models close to four months behind OpenAI and Anthropic

3 Key Points

  1. What happened

    Analysts estimate China's best models are now just four months behind OpenAI and Anthropic, versus seven months at the start of the year, and Chinese models went from 1.2% of token traffic in 2024 to over half by summer 2026.

  2. Why it matters

    The funding gap matters more than the technology gap—Chinese startups received barely a tenth of the U.S. total, and state- and venture-funding channels remain too weak for the next wave.

  3. What to watch

    Closing the performance gap increasingly depends on bulking up in-house computing capacity, but China's AI infrastructure spending remains a fraction of the U.S.'s. Watch whether Hong Kong's IPO pipeline—more than 430 applicants in the second half of 2026—becomes the main funding route.

WHO IT HITSChinese AI founders and their early-stage investors face a capital squeeze as they build frontier models, while enterprise customers and cloud providers relying on Chinese open-weight models may benefit from continued cost advantages.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

China's AI startups have narrowed the quality gap faster than many expected. Moonshot's Kimi K3, described as the world's largest open-weight model, has approached the performance of America's frontier systems, and analysts now put the best Chinese models just four months behind the most sophisticated releases from OpenAI and Anthropic—compared to seven months at the start of the year. Usage has shifted too: Chinese models went from 1.2% of token traffic in 2024 to more than half of the total by the summer of 2026.

But the body argues the bigger constraint is capital, not capability. Policy-driven funds tend to prioritize later-stage startups, early-stage venture capital is only just recovering from a three-year fundraising drought, and state banks are absorbing rising non-performing loans elsewhere on their books, which could weaken overall credit supply. Meanwhile Chinese enterprise software firms sell mainly into the domestic market, limiting their revenue base, while U.S. rivals have a global customer base, stronger brand recognition and deeper R&D budgets.

The outcome hinges on whether new funding channels can fill the gap. Hong Kong's capital markets are positioned as one of the few channels still capable of moving global capital toward Chinese enterprise at scale, with more than 430 applicants in the IPO pipeline in the second half of 2026, and private credit is another route—though these loan providers tend to prioritize bigger or established companies. Whether Chinese founders can use public listings, private credit, customer revenue sharing or their equity as collateral may determine whether they keep pace, or fall further behind.

FAQ
How much funding did U.S. AI companies get compared to China?
Between 2023 and 2026, venture funding into U.S. AI companies topped $380 billion. China's start-ups received barely a tenth of that figure, according to Boston Consulting Group.
Which Chinese AI startups went public?
Zhipu AI and MiniMax beat OpenAI and Anthropic to the public markets, raising $558 million and $620 million respectively in their Hong Kong IPOs in January, despite heavy over-subscription.
Why are Chinese AI startups listing earlier than before?
Many Chinese technology startups are choosing to list earlier than the previous generation of companies did because they lack an alternative.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleNVIDIA's P100-to-Rubin gain: same decade, two numbers, one migrating bottleneck