MANY MINDEDAlibaba previews a 2.4-trillion-parameter Qwen, claims near-frontier rank
Alibaba has previewed its next-generation model, Qwen3.8, and characterized it as second only to Anthropic's Claude Fable 5. The preview build, Qwen3.8-Max-Preview, carries 2.4 trillion parameters and is the strongest model in the Qwen family. Alibaba made it available on its Token Plan subscription and on its Qoder and QoderWork agentic platforms, announcing it in an official post on X on Sunday, July 19, 2026, during the World AI Conference in Shanghai.
Parameters are a rough measure of a system's complexity, and higher counts generally, though not always, track with stronger models. For scale, Beijing-based Moonshot AI unveiled Kimi K3 at 2.8 trillion parameters days earlier, on Thursday, described at the time as the world's largest open-source model. Alibaba positions Qwen3.8 as comparable to leading frontier systems and behind only Fable 5, which the source identifies as Anthropic's most capable widely released model.
Another large provider crowding the frontier matters for abundance in knowledge. The closer capable models cluster near the top, and the more they come from cost-competitive suppliers, the harder it is for any single vendor to hold a monopoly on advanced reasoning, which tends to push the price of capability down and its availability up.
The caveats are important. The second-only ranking is Alibaba's own claim, not an independent benchmark, and this is a preview rather than a full release. Parameter count is only a loose proxy for capability. Worth watching: third-party evaluations and whether the shipped model holds the position Alibaba is claiming for it.
Source: SCMP Tech