Alibaba previews Qwen 3.8 Max, claims second place behind Claude Fable 5

Alibaba used the World AI Conference in Shanghai (July 19) to preview its new flagship, Qwen 3.8 Max, claiming a 2.4-trillion-parameter multimodal architecture and a benchmark ranking second only to Anthropic's Claude Fable 5. Early hands-on reports indicate the model posts faster execution times than most competitors in real-world coding tasks, positioning speed as a headline differentiator. It's accessible via Alibaba's Token Plan, with open weights promised 'soon' but no firm date — a hedge that undercuts the open-source framing.

The competitive subtext is pointed. Alibaba banned employees from using Anthropic's Claude Code on July 6, classifying it as high-risk software and replacing it internally with its own Qoder tool — so a Qwen model benchmarking itself against Claude Fable 5 is as much a statement of rivalry as of capability. Separately, Qwen was approved to power Apple Intelligence in China (July 16), a major distribution win.

Alibaba also shipped Qwen-Audio-3.0-TTS this week, a text-to-speech model in Flash (real-time) and Plus (high-quality) flavors with fine-grained inline emotion tags — evidence the Qwen family is broadening across modalities, not just chasing a single flagship number.

The skeptical read: self-reported 'second place' rankings and 2.4T parameter counts need independent verification, and 'open weights soon' has become a familiar deferral. Still, Qwen 3.8 Max lands in the middle of the week's Chinese-open-weight momentum story alongside DeepSeek and Kimi K3. What to watch: whether the open weights actually drop, and how the Apple Intelligence integration in China performs at scale.