Alibaba has announced Qwen3.8 Max, a flagship language model containing 2.4 trillion parameters. The new system is capable of operating autonomously for more than 10 days to complete complex software engineering and microchip design tasks without human oversight. Developers can access the model now through Qwen Studio and API channels with prices starting at 2 dollars per million input tokens.
During testing, developers assigned the system a complex programming project from scratch and allowed it to run independently. The model successfully managed the design workflow over a 10 day autonomous cycle. Promotional footage from the company shows scenarios where human specialists rest while the background software handles tasks in chip architecture, bioengineering, economics, and legal design. The system manages these tasks through native multimodality, autonomous programming, and closed loop adaptive learning.
Alibaba claims that laboratory benchmarks place the model on par with top tier industry systems like Claude Fable, Claude Opus 4.8, and GPT 5.6 Sol. The company even reports that the architecture outperforms some competing options from Anthropic and OpenAI in specific tests. Running the system via the API costs 2 dollars per million input tokens and 6 dollars per million output tokens.
Local deployment is also on the horizon for independent developers who want to avoid cloud costs. Alibaba plans to release the model weights for Qwen3.8 Max and the smaller Qwen3.8 27B next week. This release will allow companies to host the systems locally on their own servers for custom projects.
