Qwen3.8-Max-Preview - Alibaba Tongyi's Latest Generation Base Model

What is Qwen3.8-Max-Preview

Qwen3.8-Max-Preview is the preview version of the latest generation base large model launched by Alibaba Tongyi Qianwen, with 2.4T parameters, and has been launched on Qoder, Token Plan, and the QoderWork platform. The model demonstrates strength comparable to world-leading models in complex multi-Agent long-horizon tasks such as full-stack development, data analysis, and Office workflow automation.

Main Features of Qwen3.8-Max-Preview

Ultra-Long Context Processing: Supports three context window configurations of 200K / 400K / 1M, meeting the needs of long-document understanding and long-horizon dialogue.
Deep Thinking Mode: Built-in thinking mode toggle, which can autonomously switch between deep reasoning and fast response based on task complexity.
Multi-Agent Collaboration: Supports complex task decomposition and multi-agent collaboration, covering scenarios such as full-stack development, data analysis, and office automation.
Full-Modality Capabilities: Under a unified subscription system, full-modality AI capabilities such as text, speech, and images can be invoked.
Compatible with Mainstream Tools: Supports OpenAI / Anthropic protocols, and can seamlessly integrate with Cursor、Windsurf and other mainstream Coding/Agent tools.

Technical Principles of Qwen3.8-Max-Preview

Ultra-Large-Scale MoE Architecture: Adopts a mixture-of-experts architecture with a total of 2.4T parameters, unleashing strong performance while maintaining inference efficiency through a sparse activation mechanism.
Continuous Evolutionary Training: The model is continuously iterated and optimized on a daily basis, achieving significant improvements over Qwen3.7-Max in internal real-task evaluations.
Long-Horizon Dependency Modeling: By extending the context window to the million level, it enhances semantic understanding of ultra-long texts and the ability to associate information across paragraphs.

How to Use Qwen3.8-Max-Preview

The Qwen3.8-Max preview model has been first launched onAlibaba Token Plan、QoderandQoderWork。

Core Advantages of Qwen3.8-Max-Preview

Top-Tier Performance: Officially claimed to be possibly the most powerful model apart from Fable 5, matching world-leading levels in complex real-world tasks.
Open-Source Commitment: After the preview version is experienced first, the official version will be released and open-sourced, balancing commercial applications and the community ecosystem.
Flexible Subscription: Token Plan offers multiple tiers for personal and team versions, supporting concurrent operation of 1 to 8 Agents.

Comparison of Qwen3.8-Max-Preview with Similar Competitors

Comparison Dimension	Qwen3.8-Max-Preview	Kimi K3
Parameter Count	2.4T（MoE）	2.8T（MoE）
Context Window	Up to 1M	Up to 1M
Thinking Mode	Supports toggle switching	Supports deep thinking
Pricing Strategy	Limited-time 10% of original price, 2% of original price at night	Standard Credits billing
Open-Source Plan	Official version to be open-sourced soon	Not open-sourced
Multi-Agent Concurrency	Supported (Token Plan up to 8)	Supported
Access Method	Qoder / Token Plan / API	Kimi Client / API

Application Scenarios of Qwen3.8-Max-Preview

Complex Software Development: Use multi-Agent collaboration to complete full-stack project planning, code writing, debugging, and documentation generation.
Massive Data Analysis: Leverage a 1M context window to process ultra-long reports and log files, extract key insights, and generate visualized conclusions.
Intelligent Office Automation: Connect Office workflows and automatically complete tasks such as email writing, PPT creation, and meeting minutes organization.
In-Depth Study of Long Documents: Perform summarization, question answering, and cross-chapter关联 analysis on ultra-long texts such as academic papers, legal contracts, and technical manuals.
AI-Assisted Creation: Serve as the underlying model in Coding/Agent tools, providing developers with real-time code completion and architecture suggestions.