China delivers a one-two punch to America’s AI dominance
The Verge Robert Hart ● Covered by 75 sources
China's Moonshot and Alibaba just dropped two huge new AI models, Kimi K3 and Qwen3.8, claiming they rival OpenAI and Anthropic's best. Unlike the top US labs, both are being made openly available, and cheaply too.
Based on reporting by The Verge, Robert Hart — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Beijing-based Moonshot AI kicked things off on Friday with Kimi K3, a model it says outperforms nearly every American system on its own internal tests, losing out only to OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5 — and even beating those two on select benchmarks. Alibaba wasn't far behind, previewing Qwen3.8 over the weekend and calling it one of the most powerful models available today, second only to Fable 5.
What makes this pairing notable isn't just the bragging rights. It's the strategy. Moonshot describes Kimi K3 as the largest open-source AI system in the world, built on 2.8 trillion parameters. Alibaba puts Qwen3.8 at 2.4 trillion parameters and says it's still evolving. Neither OpenAI nor Anthropic reveal parameter counts for their flagship models, so these are rare, specific numbers in an industry that usually deals in vague superlatives. Parameter counts aren't a perfect proxy for capability, but they're a signal of scale, and these are big signals.
There's a catch, though. Neither model is fully out yet. Moonshot says complete model weights will land on July 27th, a week out from the announcement. Alibaba is vaguer, saying Qwen3.8 is going open-weight soon. Until independent testers get their hands on both, the claims from Moonshot and Alibaba are just that — claims, made by the companies with every incentive to look strong.
Still, the timing and confidence echo what happened when DeepSeek released a low-cost model last year that matched leading US systems and rattled the industry. This latest one-two punch raises the same uncomfortable question for American AI labs: does pouring enormous money into chips and data centers actually buy a lasting edge, if Chinese competitors can approach that frontier while spending less?
And there's an irony sitting underneath all of it. Washington has leaned hard on export controls to slow China's access to advanced chips, and reportedly pushed Anthropic to pull its most capable model from the market over fears it could help rivals catch up. Meanwhile, Moonshot and Alibaba are doing the opposite of hoarding — they're handing their most advanced models to anyone who wants to download, tweak, and build on them. Whether the performance claims hold up under scrutiny is still unknown. But the pattern is becoming familiar: America restricts, China opens, and the gap between the two keeps looking smaller than expected.
My take — AI-written commentary, not fact-checked reporting
The real story here isn't whether Kimi K3 or Qwen3.8 actually beats GPT-5.6 Sol on some leaderboard — that's PR until independent testers get the weights. The real story is that US labs keep locking their best work behind APIs while Chinese firms keep giving theirs away, and that's a strategy difference, not a capability gap. Export controls can slow chip access, but they can't stop someone from just publishing better ideas for free, and that's the part Washington seems slow to reckon with.
Read more about this at: The Verge