China's WAIC Week Puts Two Frontier Models on the Board — Both Chasing Open Weights
Alibaba previewed Qwen3.8-Max at a claimed 2.4 trillion parameters while Moonshot AI's Kimi K3 hit frontier benchmarks and promptly ran out of subscription capacity. Both are gesturing at open weights, and that's the part Western labs should be watching.
The World Artificial Intelligence Conference in Shanghai has become the venue where China's labs announce, and this week they announced loudly. Alibaba previewed Qwen3.8-Max and Moonshot AI shipped Kimi K3, with both reaching what observers described as frontier-level performance, as summarized by @shushant_l in a weekly roundup covering July 13 to 19. The cadence is the story as much as the capability: two credible frontier releases inside a single conference cycle.
Qwen3.8-Max is the headline number. @TraffAlex reported the model carries 2.4 trillion parameters and is positioned directly against Anthropic's Claude Fable 5. Parameter counts are a blunt instrument for judging a model, and a 2.4T figure tells you more about ambition and compute access than about real-world quality. But the framing — an explicit challenge to a specific Western frontier model — is deliberate. Alibaba is not pitching a fast-follower. It is pitching a peer.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.