Alibaba unveils Qwen 3.8 with a self-reported frontier claim.

What happened.

Alibaba’s Qwen team unveiled Qwen 3.8, an open-weight multimodal model that it describes as having 2.4 trillion parameters. The team says the model rivals leading systems and is “second only to Fable 5.” A preview is available now.

The release arrives alongside another Chinese open-weight launch in the supplied record, Moonshot’s Kimi K3. Together, the two announcements point to continued pressure from Chinese open-weight releases that are being compared with Anthropic’s flagship.

Why it matters.

For teams choosing models, Qwen 3.8 may add another option to evaluate, particularly where open-weight availability matters. The related record also cautions that capability can be uneven across workloads: it characterizes these models as strong on frontend and coding work but weak on hard math and reasoning.

That makes model selection a workload-specific question rather than a conclusion that follows from a headline comparison. The supplied evidence supports evaluation, not a general performance verdict.

What to watch next.

The most useful receipt will be independent testing of Qwen 3.8, including comparisons on the actual tasks teams intend to run. Until that appears, Alibaba’s ranking claim, parameter count, and benchmark framing should be treated as reported rather than verified.