Ask the question out loud this week and it does not sound crazy anymore. Don’t intend for this to be clickbait, but trying to get a genuine point across.

In the past two days, China's Moonshot shipped Kimi K3, the largest open model ever built and the third most capable in the world. Today its president, Xi Jinping, is personally opening Shanghai's World AI Conference, the first time a Chinese head of state has done so, to pitch a China-hosted body that would write the global rules for AI. And on the exact date its own flagship was supposed to prove itself, Google's Gemini 3.5 Pro is missing its deadline again.

So has the lead actually shifted? The honest answer is more interesting than the headline, and it cuts both ways. Here are the receipts.

Start with the models. Moonshot's Kimi K3, out July 16, is a 2.8-trillion-parameter open model, the largest ever released, and it benchmarks third in the world behind only Claude Fable 5 and GPT-5.6 Sol, with open weights promised July 27. It is not a one-off, either: Chinese open-weight models now hold four of the top five global open slots across DeepSeek, Kimi, GLM, and Qwen, Wall Street analysts have started formally recommending them to clients on price and performance, and Meituan's 1.6-trillion LongCat was trained end to end without a single Nvidia chip, straight through the export controls.

Then there is the politics. Today Xi Jinping is personally opening Shanghai's World AI Conference, running through July 20 with more than 140 forums and 1,100 exhibitors, and doubling this year as a high-level meeting on global AI governance. Analysts expect him to formally propose a World AI Cooperation Organization, an international body headquartered in Shanghai to convene the global rulebook for AI, at a moment when Washington has no equivalent proposal on the table. A head of state personally opening an AI trade show would have been unthinkable two years ago. Beijing is treating AI leadership as a top-tier national priority and saying so on the world's biggest stage.

While China shipped and convened, the West's most anticipated model spent the day stuck backstage. Google's Gemini 3.5 Pro was expected to reach general availability today, July 17, the date every outlet has been circling for weeks. As of now, Google has not confirmed a launch, there is still no gemini-3.5-pro model ID on its public API pages, and the smart money expects another slip.

The backstory is rough. Sundar Pichai told the I/O crowd in May to wait until next month, June came and went, and Google then scrapped the model's entire base architecture for a ground-up rebuild after early testers flagged failures in math reasoning and SVG generation. This would be the third missed deadline, and prediction markets have priced it in: a Polymarket contract with more than $320,000 in volume now puts a July 31 delivery at 81 percent, with a separate market leaning toward August 7. Google is the only major frontier lab without a 2026 flagship shipped on its own promised date, and it is happening the same week China put both its newest model and its president on the board.

Here is the part the hype leaves out, and the reason the headline has a question mark.

The United States still holds the very top of the frontier. Claude Fable 5 and GPT-5.6 Sol both rank above Kimi K3, and Fable 5 still leads the hardest coding benchmarks. American labs also still dominate the money and the enterprise: the vast majority of global AI venture funding this year went to US startups, and the biggest companies still build their production stacks on Anthropic, OpenAI, and Google. K3 is also pricier and slower than the cheap Kimis that built the brand.

So the honest read is not that China has seized the lead outright. It is that China now owns the open-model frontier, is closing fast on the closed one, and out-executed the US on both shipping and diplomacy in a single week when the American flagship stalled. That is a narrower claim than the headline, and a more uncomfortable one for US labs, because it is true.

For builders, the move is to stop treating open Chinese models as second-tier. Put Kimi K3 and its peers in your eval now, circle July 27 for the weights, and price out what frontier-adjacent quality at open-weight cost does to your stack. The margin you have been relying on, that the good models only come from US APIs, is thinner this week than it was last week, and the leaderboard's top line hides how fast the rest of it is rising.

The stories moving fast beyond today's headlines:

This was the week the "is China ahead?" question stopped sounding crazy. Moonshot shipped Kimi K3 on July 16, a 2.8-trillion-parameter open model that is the largest ever released and benchmarks third in the world behind only Fable 5 and GPT-5.6 Sol, with open weights due July 27. It is not a one-off: Chinese open models now hold four of the top five global open slots, Wall Street has started recommending them on price and performance, and Meituan's LongCat was trained without a single Nvidia chip. On top of that, Xi Jinping is personally opening Shanghai's World AI Conference today, the first Chinese head of state ever to do so, and is expected to propose a China-hosted body to write the global AI rulebook while Washington has nothing comparable on the table. Meanwhile the West's most anticipated model stalled: Google's Gemini 3.5 Pro missed its July 17 deadline, still has no model ID on Google's API, and prediction markets now put another slip to July 31 at 81 percent, leaving Google the only major lab without a 2026 flagship shipped on its own promised date. But the honest scoreboard keeps the question mark: the US still holds the top two models and the vast majority of AI capital and enterprise adoption, so the real story is not that China has taken the lead, it is that China now owns the open frontier, is closing fast on the closed one, and out-shipped and out-maneuvered the US in a single week. For anyone watching the race, the margin the US has leaned on is thinner today than it was seven days ago.

Stay building. 🤖