Alibaba's Qwen 3.8 open-weight model claims #2 ranking, eyes Kimi K3

Alibaba has announced Qwen 3.8, an open-weight large language model built with 2.4 trillion parameters. According to the Qwen team, the model trails only Fable 5 in performance and is positioned to outperform Qwen 3.7-Max, Alibaba's prior flagship. The team highlights strength in coding and complex productivity tasks, including full-stack development, data analysis, and office workflows. Developer Shuai Bai notes that Qwen 3.8 marks a milestone for the team: it is their first multimodal model with more than 1 trillion parameters, capable of processing images, videos, and documents in addition to text. However, no benchmark results have been published yet.

Access is rolling out in stages. A preview version is immediately available through Alibaba's Token Plan, Qoder, and QoderWork platforms at 10 percent of the standard price. Full open weights are expected to arrive soon, though Alibaba has not announced a specific date. This tiered release strategy allows developers to evaluate the model early while the team prepares the open-source version for broader distribution.

The move appears strategically timed against Moonshot AI's Kimi K3. Moonshot has built significant momentum, reaching $300 million（約480億円） in annual recurring revenue as of June, and has promised to release open weights. However, current access remains gated through Moonshot's chat app and API—a restriction that converts user interest into paying customers. By releasing Qwen 3.8 as open weights now (or imminently), Alibaba is pursuing a contrasting strategy: transparency and developer access rather than access control. The competitive dynamic is sharpened by Moonshot's plans to go public within six months, which may accelerate its monetization timeline and make Alibaba's open alternative more attractive to developers seeking unrestricted model deployment.

Alibaba's Qwen 3.8 enters a competitive landscape shaped by Moonshot AI's Kimi K3, which has built significant traction by restricting access through its proprietary chat app and API—a strategy that converts interest into paying customers. Moonshot reached $300 million（約480億円） in annual recurring revenue in June and plans to go public as soon as six months, signaling that the model's gated-access approach has been commercially successful. Qwen 3.8's open-weight strategy represents a directly opposing model: by releasing the model weights to the community and offering a discounted preview now, Alibaba appears to be betting that developer accessibility and transparency will disrupt Moonshot's momentum.

The timing and positioning are deliberate. Qwen 3.8's multimodal capabilities—the ability to process images, videos, and documents across more than 1 trillion parameters—address the same productivity and full-stack development workflows that Kimi K3 targets. Alibaba's claim that Qwen 3.8 ranks second only to Fable 5 (though with no benchmarks published yet) frames the release as a frontier-tier challenger. The gap between a preview at discounted price and the arrival of full open weights suggests Alibaba is managing rollout to gather feedback while maintaining momentum.