Alibaba Unveils Qwen3.8, a 2.4 Trillion-Parameter AI Model, Launches Preview with Open-Weight Commitment

Alibaba's Qwen team has announced the launch of Qwen3.8, a new large multimodal language model boasting 2.4 trillion parameters. The company has made a preview version, Qwen3.8-Max-Preview, immediately available on its Token Plan subscription service, as well as its Qoder integrated development environment (IDE) and QoderWork desktop agent platforms. Alibaba also confirmed that the full Qwen3.8 model is slated for an open-weight release in the near future, marking a significant strategic shift for its flagship models.

The introduction of Qwen3.8 positions Alibaba as a formidable competitor in the rapidly escalating race for advanced AI models, particularly in the open-weight domain. This release follows closely on the heels of other multi-trillion parameter announcements from Chinese AI developers, highlighting a renewed push for innovation and accessibility in the sector.

1. Qwen3.8: Scale, Multimodality, and Performance Claims

Qwen3.8 stands out due to its substantial scale, incorporating 2.4 trillion parameters, making it one of the largest models publicly known. The model is also designed to be multimodal, capable of processing and generating content across various input types including text, images, and video. It is the first multimodal model developed by Alibaba's Qwen team to exceed one trillion parameters.

Alibaba has stated that Qwen3.8's overall performance is comparable to leading frontier AI models currently available, asserting it is "second only to Fable 5" – a reference to Anthropic's highly capable Claude Fable 5 model. While Alibaba has yet to release independent benchmarks to substantiate these claims, the announcement underscores the company's confidence in Qwen3.8's capabilities. Anthropic's Claude Fable 5 is known for its advanced reasoning and long-horizon agentic work, featuring a 1 million token context window. Community reports suggest Qwen3.8-Max-Preview also offers a 1 million token context window. The architecture of Qwen3.8 is built on a sparse Mixture-of-Experts (MoE) design.

2. Immediate Accessibility and Future Open-Weight Release

The Qwen3.8-Max-Preview is currently accessible through several of Alibaba's AI-focused platforms. Users can engage with the model via Alibaba's Token Plan, a subscription service designed for AI programming scenarios and long-context agentic workflows. The preview is also integrated into Qoder, an agentic IDE for software development, and QoderWork, a desktop AI agent for macOS and Windows that automates complex tasks directly on a user's local machine.

To encourage adoption during the preview phase, Alibaba is offering significant discounts. Personal Token Plan users can access Qwen3.8-Max-Preview at a 90% discount on credit consumption during the day, with an additional 80% discount applied during night hours (22:00 to 08:00 SGT). The personal Token Plan offers various tiers ranging from 39 CNY (approximately $5.76 USD) to 499 CNY (approximately $73.67 USD) per month.

A key aspect of the Qwen3.8 launch is Alibaba's commitment to releasing the model with open weights "soon." This move signals a strategic shift, as previous Max-tier models in the Qwen family, such as Qwen3.7-Max and Qwen3.7-Plus, were predominantly proprietary and offered only via API. The decision to open-source Qwen3.8 is anticipated to further democratize access to advanced AI capabilities and foster innovation within the developer community.

3. Competitive Landscape and Strategic Implications

The debut of Qwen3.8 comes amidst intense competition within the AI industry, particularly among Chinese developers. Just days prior, Beijing-based Moonshot AI unveiled its Kimi K3 model, featuring 2.8 trillion parameters and also committed to an open-weight release. This rapid succession of multi-trillion parameter model announcements highlights a vigorous race among Chinese AI labs to push the boundaries of large language models and establish leadership in the open-source AI arena.

Alibaba's strategic decision to offer Qwen3.8 with open weights aims to reclaim a prominent position in the open model narrative, especially as developers increasingly seek access to powerful, customizable models. The company's dual-track strategy of offering both subscription-based services and open-source models is designed to cultivate user payment habits while enabling broad developer engagement. The enhanced capabilities of Qwen3.8 are expected to significantly impact areas such as full-stack development, data analysis, and professional office workflows, providing advanced tools for a wide range of users and enterprises.

Frequently Asked Questions

What is Qwen3.8?

Qwen3.8 is Alibaba's latest large multimodal language model, featuring 2.4 trillion parameters and designed for advanced reasoning, code engineering, and professional office automation tasks.

Is Qwen3.8 available now?

A preview version, Qwen3.8-Max-Preview, is currently available on Alibaba's Token Plan, Qoder, and QoderWork platforms. The full Qwen3.8 model is expected to have an open-weight release soon.

How does Qwen3.8 compare to other AI models?

Alibaba claims that Qwen3.8 is one of the most powerful models available today, comparable to leading frontier AI models, and is positioned as "second only to Fable 5" (Anthropic's Claude Fable 5).

What are the main features of Qwen3.8?

Qwen3.8 is a multimodal model capable of processing text, images, and video. It is built on a sparse Mixture-of-Experts (MoE) architecture and is designed to excel in complex, long-horizon tasks such as full-stack development, data analysis, and office workflows.

Sources

* Alibaba_Qwen. (2026, July 19). *Qwen3.8 is launching and going open-weight soon!🌐 With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5.* [X post]. https://x.com/Alibaba_Qwen/status/2078759124914098291 * Alibaba Cloud's Qwen Announces Open 2.4T Parameter Large Model Qwen3.8 - Phemex News. (2026, July 19). *Phemex News*. * Alibaba Launches Qwen 3.8 With 2.4 Trillion Parameters, Claims Near-Frontier Performance. (2026, July 19). *AIHubMix*. * Qwen3.8-Max-Preview Launch — 90 Percent Off, Night-Time 98 Percent Off - Qoder Docs. (2026, July 19). *Qoder Docs*. https://qoder.ai/docs/qwen3.8-max-preview-launch * Qwen - Wikipedia. (n.d.). Retrieved July 19, 2026, from https://en.wikipedia.org/wiki/Qwen * Introducing Claude Fable 5 and Claude Mythos 5 - Claude Platform Docs. (2026, July 1). *Anthropic*. https://www.anthropic.com/news/claude-fable-5 * Alibaba says newest Qwen AI model is second only to Anthropic's Claude Fable 5. (2026, July 19). *South China Morning Post*. * Token Plan - MiniMax API Platform. (n.d.). Retrieved July 19, 2026, from https://mimax.ai/token-plan * Introducing QoderWork: An AI Assistant That Actually Gets Work Done | Qoder. (2026, February 12). *Qoder*. https://qoderwork.org/blog/introducing-qoderwork * Qoder is an agentic IDE for real software development. (n.d.). *Product Hunt*. Retrieved July 19, 2026, from https://www.producthunt.com/products/qoder * Alibaba Launches Token Plan Individual Tier, Debuts 2.4T-Parameter Qwen3.8-Max Preview - BigGo Finance. (2026, July 19). *BigGo Finance*.