Alibaba Previews Qwen3.8-Max, a 2.4 Trillion-Parameter Multimodal Model, Days After Moonshot’s Kimi K3 Open…
What happened
Alibaba’s Qwen team released a preview of Qwen3.8-Max-Preview, a multimodal model with 2.4 trillion parameters using a mixture-of-experts (MoE) architecture. Alibaba positions it as the second-largest after Fable 5. The preview is available now on Token Plan, Qoder, and QoderWork platforms at 10% of the usual price. However, Alibaba has not yet shared any benchmark results, a detailed model card, official licensing, per-token costs, or exact active parameter counts.
Why it matters
Qwen3.8-Max marks Alibaba’s attempt to scale up AI model size aggressively, keeping pace with other massive models in the trillion-parameter range. For builders and enterprises, the lower pricing preview could lower barriers to trialing massive models for multimodal tasks like combining text, images, or other data types. At the same time, the lack of transparency on anything beyond parameter count raises doubts about real-world performance, costs, and use case fit. Without benchmarks or pricing clarity, it is difficult for operators to evaluate whether this model improves on existing alternatives or just increases complexity and infrastructure demands.
What to watch next
Pay close attention to when Alibaba releases benchmark data and pricing details. Those will be crucial to understanding whether Qwen3.8-Max will shift market economics for large-scale AI. Also watch Moonshot’s Kimi K3 launch since it came just days earlier with an open-weight model, setting a direct comparison point. The effective usage, efficiency, and pricing of both will influence enterprise AI procurement decisions. Finally, assess Alibaba’s licensing model once it appears to gauge commercial risks for builders and integrators.
AI Quick Briefs Editorial Desk
