Alibaba is offering developers preview access to Qwen3.8 Max at around 10 percent of its normal rate, creating a lower-cost entry point to test one of its largest AI models. The 2.4-trillion-parameter model is currently available through selected Alibaba services rather than as a downloadable release.
The pricing approach arrives as competition among major AI developers in China increasingly involves access terms and model openness, not only parameter counts. Alibaba has said the model weights will eventually be released, although it has not provided a specific launch date.
Preview access comes before an open-weight release
Developers can access Qwen3.8 Max through Alibaba’s Token Plan subscription service and its AI coding tools, Qoder and QoderWork. The temporary discounted rate is intended to let users explore the model’s capabilities without immediately paying the full tariff.
Alibaba plans to make Qwen3.8 Max an open-weight model. This would allow developers to download the trained weights, run the model on their own infrastructure, and adapt it for their own needs.
For now, the model remains limited to Alibaba-provided channels. The company has only said that the weights will become available “soon,” leaving the timing of a self-hosted release unclear.
A model built for extended work
Qwen3.8 Max is positioned as an advance over Qwen3.7 Max for work requiring longer chains of reasoning and execution. Its intended uses include code generation, full-stack application development, data analysis, and office workflow automation.
Alibaba also presents the model as a foundation for AI agents that can complete a series of tasks with help from tools or other applications. The goal is to support structured work processes rather than limit the system to answering individual questions.
The model is multimodal, meaning it can process several input formats. In addition to text, it can handle images, video, and documents.
| Model | Parameters | Stated status | Core focus |
|---|---|---|---|
| Qwen3.8 Max | 2.4 trillion | Preview, weights planned | Coding, data analysis, AI agents |
| Kimi K3 | 2.8 trillion | Open-weight | Coding, reasoning, AI agents |
Large scale, but limited public evidence
Alibaba describes Qwen3.8 Max as a frontier-level AI model and places it below Claude Fable 5. That assessment is based on internal testing and has not yet been supported by independent comparisons.
The company has not published a technical report, model card, or benchmark table for public review. Without those materials, outside parties cannot measure its capabilities against competing systems on equal terms.
Shuai Bai, a Qwen developer, described Qwen3.8 Max as his team’s first multimodal model to exceed one trillion parameters. The statement was reported by Bloomberg and cited by tekno.kompas.com.
Parameter count reflects a model’s capacity to learn patterns during training and use them when generating responses. It does not, by itself, determine the final quality of an AI model.
China’s race for larger and more open models
Alibaba’s announcement came days after Moonshot AI introduced Kimi K3, a 2.8-trillion-parameter model presented as the largest open-weight model in its category. Kimi K3 is also aimed at coding, reasoning, and AI agent tasks.
The two launches show how Chinese technology companies are pursuing larger models while also emphasizing more open availability and lower costs. For Alibaba, Qwen is part of a wider effort to strengthen its AI and cloud computing business across enterprise services, e-commerce, and internet platforms.
