Enough Being Surprised by Chinese AI

Last week, two Chinese companies at once showed off AI models that, by their own claims, can compete on equal terms with the best that OpenAI and Anthropic have to offer. The reaction was predictable: markets wobbled, commentators declared Silicon Valley to be in shock, and politicians once again started talking about an arms race and a “wake-up call.” Headlines were peppered with phrases like “caught off guard,” “a breakthrough that's crashing markets,” and even “America's Sputnik moment.”

The Verge's Robert Hart notes that the only truly surprising thing here is that anyone was surprised at all. For years we've been warned that China is catching up — and now that the moment seems to have arrived, the world feigns astonishment. Six of the top ten models by token consumption on OpenRouter's leaderboard are Chinese. The quality gap has long been narrowing, and Chinese systems are also markedly cheaper: American startups are reportedly switching to them more and more as prices climb at domestic providers.

What exactly they showed

Beijing startup Moonshot AI released its flagship Kimi K3, claiming it trails only OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5. The pricing is aggressive — $15 per million output tokens, versus roughly $30 for GPT-5.6 Sol and $50 for Fable 5. Demand ran so high that the company temporarily paused new subscriptions. Next, Alibaba unveiled a preview of Qwen3.8, calling it “one of the most powerful models available today” and “second only to Fable 5.”

The key point: both companies promise to release models with open weights — in contrast to the closed approach of OpenAI, Anthropic, and Google.

Why this matters more than the ranking

The economics deserve a sober look. Cheaper doesn't mean better: tokens from different models aren't comparable, and companies often subsidize inference to lure customers away. There's also the unresolved dispute over whether the Chinese are training their systems on American models. But even while lagging behind the cutting edge, “good enough” and affordable models can siphon off demand, squeeze margins, and undermine the expectations propping up the future trillion-dollar IPOs of Anthropic and OpenAI — and with them a sizable chunk of the American market.

There's a security question, too. Open Chinese models grant access to capabilities in areas where American companies impose restrictions. Hart cites reports that Kimi K3 found and fixed vulnerabilities that Codex and Fable wouldn't touch because of their guardrails.

The models haven't been fully released yet, and the benchmarks should be treated with caution. But the exact ranking, the author concludes, hardly matters. Leading Chinese companies are putting out worthy rivals regularly enough that each release has stopped being a shock. If this really is a race, it's time to admit it: someone else might win — or come so close that the difference disappears.

Source: The Verge