Qwen 3.8-Max: A New Benchmark in Autonomous Coding and Coworking

Qwen has officially released Qwen 3.8-Max, the most advanced model in their lineup to date. With a staggering 2.4 trillion parameters (95 billion active), this release marks a significant leap forward in AI-assisted development. Qwen 3.8-Max not only answers complex questions but also completes multi-day projects from scratch, all without human intervention. The open-source weights for the model will be released next week, making it accessible to developers and researchers alike.

Key Improvements and Capabilities

Qwen 3.8-Max builds on the foundation of Qwen 3.5, bringing comprehensive improvements across coding, work, research, and long-horizon tasks. Here are some key highlights:

• 2.4T Parameters (95B Active): The model's massive size allows it to handle more complex and diverse tasks.
• Autonomous Coding: Qwen 3.8-Max can take on real-world projects, from an empty folder to a finished result, without human assistance.
• Self-Evolving Feedback Loops: It continuously refines its approach through feedback loops, improving over time.

To demonstrate these capabilities, Qwen tested 3.8-Max on three challenging tasks, each requiring the model to write and run code independently. One notable project was the creation of oh-my-cli, a command-line interface tool developed over 10 days with no human intervention. The entire process, from requirement gathering to final delivery, is documented in the GitHub repository qwen-code-dev-bot/oh-my-cli.

Key Implementation Details

The success of Qwen 3.8-Max in these tasks can be attributed to its robust implementation details:

• Loop Engineering Setup: The model combines an issue state machine, dispatcher, monitor, and watchdog into a single execution loop. When a new requirement is added to GitHub Issues, an agent claims it through the state machine and moves through ready → leased → active. Once the implementation is complete, end-to-end tests and CI checks are triggered, and the PR is merged after passing all checks.
• Self-Testing: Qwen 3.8-Max includes a self-testing framework that integrates user feedback, advanced community practices, and its own test results into one engineering loop. This ensures continuous improvement and reliability.

In Practice

Qwen 3.8-Max's capabilities extend beyond just coding. It can also handle long-horizon tasks, research projects, and complex workloads with greater reliability. For developers, this means:

• End-to-End Project Management: From requirement gathering to final delivery, Qwen 3.8-Max can manage the entire project lifecycle.
• Reduced Development Time: By automating repetitive tasks and providing reliable code generation, it accelerates development cycles.
• Enhanced Collaboration: The model can work alongside human developers, providing support and insights throughout the development process.

Simon Willison, a prominent developer, noted that while Qwen 3.8-Max is excellent, it has a tendency to overthink tasks by default. However, this behavior can be controlled using the reasoning_effort parameter, which allows developers to adjust the model's reasoning depth and balance cost and performance.

What to Watch

As Qwen 3.8-Max becomes more widely adopted, several areas will be worth monitoring:

• Community Adoption: How quickly and effectively the developer community integrates Qwen 3.8-Max into their workflows.
• Open-Source Contributions: The impact of open-sourcing the model's weights on research and development in the AI community.
• Performance Benchmarks: Ongoing performance benchmarks to ensure the model continues to deliver on its promises.

Qwen 3.8-Max represents a significant step forward in AI-assisted development, offering developers a powerful tool to tackle complex projects with greater efficiency and reliability. Whether you're a solo developer or part of a large team, Qwen 3.8-Max is worth exploring for its potential to transform your coding experience.