- Published on
Alibaba Previews Qwen3.8-Max, a 2.4 Trillion-Parameter Multimodal Model, Days After Moonshot’s Kimi K3 Open-Weight Launch
- Authors
- Name
- DeepHorus
Alibaba's Qwen team previewed Qwen3.8-Max-Preview on July 19, 2026, a 2.4 trillion-parameter multimodal model that it claims is "second only to Fable 5" among benchmarked systems. The announcement arrived during the World AI Conference in Shanghai, just two days after Moonshot AI released Kimi K3, a 2.8 trillion-parameter open-weight model. The timing underscores the intensifying competition in frontier AI, with Alibaba positioning its latest flagship against both domestic and global rivals. However, the preview launch lacks published benchmarks, model cards, or license details, leaving many claims unverified.
The Qwen team described Qwen3.8-Max as its first multimodal model above 1 trillion parameters, capable of processing text, images, video, and documents. Developer Shuai Bai stated it should outperform Qwen3.7-Max on coding, full-stack development, data analysis, and office workflows. The preview is available via Alibaba's Token Plan subscription at 10% of standard pricing, and it supports OpenAI and Anthropic protocol compatibility for seamless integration. Yet, Alibaba has not confirmed whether open weights will follow, as its previous two Max flagship models remained closed.
Key figures from the announcement include the claimed 2.4 trillion total parameters, though this number requires verification. Qwen3.8 uses a sparse Mixture-of-Experts architecture, meaning only a fraction of parameters activate per token. The active parameter count, which determines serving cost, remains undisclosed. For context, Qwen3.7-Max achieved a GPQA Diamond score of 92.4, SWE-bench Verified of 80.4%, and a context window of 1 million tokens at $1.25 per million input tokens—but those are predecessor numbers, not yet matched by the new model.
The implications of this preview are significant but uncertain. If the performance claims hold, Qwen3.8-Max could challenge leading models like Fable 5 in multimodal reasoning and coding tasks. The 10% pricing discount during preview suggests Alibaba aims to attract developers and enterprise customers quickly. However, the absence of independent benchmarks means customers must rely on Alibaba's internal evaluations, a risk that may slow adoption among cautious enterprises.
Industry reactions highlight both excitement and skepticism. Developers on social media praised the affordable preview pricing and protocol compatibility, which lowers switching costs. But many noted that without model cards or open weights, they cannot verify parameter counts or replicate results. The lack of a license also raises questions about usage rights, especially after Alibaba's history of keeping Max models closed. Some researchers argue that total parameter counts are misleading without active parameters, pointing to Qwen3-235B-A22B as a precedent.
Looking ahead, Alibaba plans to release open weights "soon," but no date or repository has been announced. The company's track record suggests that a closed model may remain the final form unless competitive pressure forces a change. The AI community will watch for benchmark releases and third-party evaluations in the coming weeks. If Qwen3.8-Max matches its claims, it could reshape the competitive landscape, but for now, the model remains a promising preview waiting for proof.
