Alibaba's Qwen4-Max Preview: A Million-Token Multimodal Flagship for Coding and Agents
Alibaba's Qwen team has launched a preview of Qwen4-Max, a new million-token multimodal flagship model (model ID: qwen4-max-preview) designed for coding, agentic tasks, and front-end development, accepting text, image, and video input. This initial Qwen4 generation model is available for early access via Alibaba Token Plan API, Qwen Chat, Qwen App, Qoder, QwenWork, and QwenStudio, with the preview running until November 25, 2026.
Introducing Qwen4-Max: A Multimodal Powerhouse
Qwen4-Max represents the first model in the new Qwen4 generation from Alibaba. This flagship model distinguishes itself with its multimodal capabilities, accepting diverse inputs including text, images, and video. Its substantial one-million-token context window allows for extensive processing of information, while its ability to generate up to 393,216 output tokens supports complex and detailed responses.
The model is specifically engineered to address demanding workloads such as coding, various agentic tasks, and front-end development. This focus suggests an intent to support developers and advanced users in creating sophisticated AI-powered applications and systems.
Accessing the Qwen4-Max Preview
Early access to the Qwen4-Max preview is currently available through several of Alibaba's platforms. Users can interact with the model via the Alibaba Token Plan API, Qwen Chat, the Qwen App, Qoder, QwenWork, and QwenStudio. The preview's availability under the model ID qwen4-max-preview on the Alibaba Token Plan API confirms its integration into the broader Alibaba AI ecosystem.
this preview version has a defined lifespan, with retirement scheduled for November 25, 2026. This indicates that subsequent versions or a general release of Qwen4-Max are anticipated following the preview phase.
Performance Insights from Early Tests
Initial independent evaluations conducted by StealthMark provide some early insights into Qwen4-Max's performance. These tests indicate that the Qwen4-Max preview demonstrates higher visual intelligence compared to its predecessor, Qwen3.8-Max. However, the same tests also suggest that Qwen4-Max preview exhibits lower general intelligence when directly compared to Qwen3.8-Max.
These preliminary findings highlight a potential specialization in visual processing for the new model, while also pointing to areas where its general reasoning capabilities may still be evolving or optimized differently. Further comprehensive benchmarks will likely provide a more complete picture of its overall performance across various tasks.
The Broader Qwen4 Rollout
The launch of Qwen4-Max preview marks the beginning of the broader Qwen4 rollout. Alibaba has indicated that additional, smaller variants are still to be released. These include Qwen4-27B, Qwen4-Flash, and Qwen4-Plus. The staggered release of these models suggests a strategic approach to cater to different computational needs and application scenarios, from high-performance flagship use cases to more efficient or specialized deployments.
As of this announcement, specific details regarding open weights or per-token pricing for Qwen4-Max or its upcoming variants have not yet been disclosed. This information is typically provided closer to a general availability release.
Conclusion
Alibaba's Qwen4-Max preview introduces a significant advancement in multimodal AI, particularly for coding and agentic applications, with its one-million-token context window and diverse input capabilities. While early tests show promising visual intelligence, further development and evaluation will clarify its full potential. The ongoing rollout of the Qwen4 series, with more variants expected, positions Alibaba to expand its offerings in the competitive AI landscape. Users interested in leveraging advanced multimodal models for development tasks should explore the available preview access points.
Sources
- stealth-model/Qwen3.8-Flash-Next__Qwen4-news · Hugging Face
- Duplicate from stealth-model/Qwen4 · stealth-model/Qwen3.8-Flash-Next__Qwen4-news at 4cc6677
- oh-my-pi/packages/catalog/CHANGELOG.md at main · can1357/oh-my-pi · GitHub
- qwen4-max-preview: server-side Chinese tool-URL rule (routify) leaks into context, model flags it as prompt injection and comments on it · Issue #17086 · QwenLM/qwen-code · GitHub
Recommended AI tools
Adobe Photoshop
Design
Create, edit, and design with industry-leading AI-powered image innovation.
Codex
Code Assistance
Build anything with Codex
Wan
Video Generation
AI Video Creation. Realism. Audio. Control.
Adobe Firefly
Image Generation
Create your way with Adobe Firefly—AI for every creative vision.
AutoGPT
Productivity & Collaboration
Your team of AI experts
Apify
Data Analytics
Automate Anything
About the Author

Albert Schaper is a co-founder of Best-AI.org. He focuses on product strategy, AI adoption, practical tool selection, and educational content that helps users compare AI products with clearer context.
More from AlbertWas this article helpful?
Found outdated info or have suggestions? Send us a note.