Alibaba previews Qwen 3.8 Max, a 2.4T-parameter model it claims trails only Claude

Alibaba unveiled a preview of Qwen 3.8 Max, a 2.4-trillion-parameter multimodal mixture-of-experts model it claims trails only Anthropic's Claude (Fable 5) across benchmarks. Alibaba touts advanced coding, reasoning, and agentic capabilities, and reports faster execution times in real-world coding tests versus several competitors. The preview is accessible through Alibaba's Token Plan subscription and its Qoder and QoderWork agentic platforms.

The launch is part of a broader intensification of China's frontier-model race — arriving alongside Moonshot AI's Kimi K3 and DeepSeek's V4 — and coincides with China's July 15 anthropomorphic-AI rules forcing removal of AI-companion features. Qwen is being integrated across Alibaba Cloud, positioning it as the enterprise backbone for Chinese and increasingly global customers.

Competitively, Qwen 3.8 Max's speed claims and open-model heritage feed directly into the week's narrative of Chinese models rattling US Big Tech and triggering tech-stock jitters. Alibaba's Qwen team separately shipped Qwen-Audio-3.0-TTS, adding to its rapid release cadence.

Skeptics will note the 'second only to Claude' claim is self-reported on a preview model, and parameter count alone doesn't guarantee real-world quality. The 2.4T MoE figure also invites scrutiny on active-parameter efficiency. Watch for independent benchmarks, general availability, and whether Qwen's speed advantage holds under production loads against Opus 5 and GPT-5.6.