Alibaba Cloud announces Qwen4, offers Qwen4-Max preview through Token Plan and four Qwen services
Alibaba Cloud announced the Qwen4 AI model on November 11 (Japan time) and is offering a preview of Qwen4-Max through its Token Plan subscription. Early access is also available in QwenWork, Qoder, Qwen APP and QwenStudio. Alibaba claims gains in knowledge, real-world understanding, coding and autonomous task execution.
Alibaba Cloud announced its Qwen4 AI model on November 11 (Japan time) and is offering a preview of Qwen4-Max through its own AI services, according to Japanese outlet ITmedia AI+. The announcement was made on the official Qwen account on X (@Alibaba_Qwen).
What Alibaba says
Alibaba claims Qwen4 improves in four areas:
- Knowledge
- Understanding of the real world
- Coding performance
- Ability to complete tasks autonomously
The company says the model is "designed to handle real-world tasks, from complex coding to knowledge work, front-end building and professional software workflows." These are Alibaba's own claims. No benchmark results have been published alongside them.
How to access the preview
The Qwen4-Max preview is available through Token Plan, Alibaba Cloud's subscription service. Alibaba also says early access is available in four of its services:
- QwenWork
- Qoder
- Qwen APP
- QwenStudio
Specifications
ITmedia reports that details of the model are unclear. Parameter count, context window, training cutoff date, benchmark scores and pricing per 1M tokens have not been disclosed. The article (written in Japanese) describes Qwen4-Max as a preview version, not a general release.
What this means
The Qwen4-Max release is a preview, and it is delivered through Alibaba's own channels: a subscription plan and four named services. Developers and teams who want to test it can do so through Token Plan or the early-access services listed above.
Alibaba's stated emphasis is on coding, front-end work, knowledge work and software workflows, plus autonomous task execution. That positions Qwen4-Max for practical engineering and productivity use, but without published benchmarks or specifications, those claims cannot yet be compared against other models. Evaluation will depend on hands-on testing in the preview and on any technical details Alibaba releases later.
Source: ITmedia AI+ (Japanese-language), which cites Alibaba's announcement on X.
Related Articles
Cactus Compute lists Mirage, a vision model for image understanding and structured extraction, on Hugging Face
Cactus Compute has published Mirage on Hugging Face, a vision model for image understanding and structured extraction. It installs through the cactus-compute Python package, which downloads and caches the engine and weights on first use.
Baidu releases ERNIE 5.5 Flash Preview: 280B parameters, 14B active, 1M context, $0.28/1M input
Baidu has released ERNIE 5.5 Flash Preview, a model with 280 billion total parameters, 14 billion active per token, and a 1 million token context window. It is listed at $0.28 per 1M input tokens and $1.13 per 1M output tokens on OpenRouter via Baidu Qianfan.
Baidu releases ERNIE-Light-DT-1.0 with 1.0M-token context at $1.13 input, $3.94 output per 1M tokens
Baidu has released ERNIE-Light-DT-1.0 through Baidu Qianfan, with a 1.0M-token context window and pricing of $1.13 input and $3.94 output per 1M tokens. The model is based on open-source GLM models and uses Baidu Sparse Attention (BSA), which Baidu says lowers KV cache storage and compute costs.
NVIDIA releases 2-billion-parameter Alpamayo 2 Edge weights on Hugging Face for autonomous driving
NVIDIA released the Alpamayo 2 Edge model weights on Hugging Face on November 10, 2026. The 2-billion-parameter model targets autonomous-driving developers working within onboard compute and latency limits, and ships with deployment recipes and evaluation tools.
Comments
Loading...