Moonshot, Alibaba release open models challenging US AI labs — SkimNews
SkimNews Take
Open-weight releases reframe the competition from frontier capability to ecosystem adoption — if Chinese models become the default substrate for fine-tuning and self-hosted deployment, closed US labs risk losing leverage at the infrastructure layer even as they retain benchmark leadership.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Moonshot AI unveiled Kimi K3 on Friday, claiming its own tests rank it above nearly every US system, trailing only OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5 — and it leads on certain benchmarks.
- Alibaba previewed Qwen3.8 over the weekend, calling it "one of the most powerful models available today" and "second only to Fable 5," Anthropic's flagship.
- Both Chinese companies are releasing their frontier models as open-weight systems, contrasting with OpenAI and Anthropic, which keep their top models proprietary; Kimi K3 has 2.8 trillion parameters and Alibaba says Qwen3.8 is 2.4 trillion, while neither US lab discloses parameter counts.
- Moonshot plans to release full model weights on July 27th; Alibaba says Qwen3.8 is "going open-weight soon" — though independent assessment of either model's true capability awaits full release.
- The releases echo DeepSeek's low-cost model from last year that rivaled leading US systems, and raise the question of whether massive US spending on chips, data centers, and training can secure a durable lead.
- Washington has used export controls to restrict China's access to advanced chips and forced Anthropic to pull its most capable system from market over fears it could help foreign competitors catch up.
Why it matters: If Moonshot's and Alibaba's claims hold up under independent testing, China's open-weight approach gives developers worldwide free access to frontier-tier models that US labs gate behind paywalls — undermining a core moat of the $300B-plus US AI investment thesis at a fraction of the compute cost.
Ask SkimNews
