Qwen3.8-Max

Qwen3.8-Max is the latest flagship from Qwen Team, officially announced on August 3, 2026. Alibaba describes it as the most powerful model in the Qwen family and says it scales the Qwen3.5 architectural foundation to 2.4 trillion parameters. The release targets a broader class of demanding work: software engineering, office productivity, scientific research, and long-duration tasks that need to end in a trustworthy result.

• A much larger flagship: Qwen Team reports 2.4 trillion parameters, making Qwen3.8-Max its largest announced model to date.
• End-to-end coding: The launch focuses on projects that start from an empty folder and continue through implementation, tests, feedback, and delivery—not just code completion.
• Multi-day autonomy: Qwen Team describes a self-evolving coding harness that ran autonomously for about 16 days, producing 265 commits, 127 pull requests, and 151 issues by July 30, 2026.
• Research execution: In another provider case study, the model reportedly worked for about 125 uninterrupted hours, wrote roughly 7,600 lines of code, performed more than 1,100 operations, and ran 33 GPU training experiments while reproducing and extending a research paper.
• Broader knowledge-work focus: Alibaba positions the model for coding, office work, research, and long-horizon projects, making it a general flagship rather than a coding-only release.
• Open-weight direction: Qwen Team says Qwen3.8-Max will be the first Max-tier Qwen model with open weights, with the weights announced for release after the hosted launch. Verify the current release and license status before planning self-hosting.

The market shift behind Qwen3.8-Max is from impressive one-shot answers to systems that can keep learning from tests, users, repositories, and experimental results. That makes the model most interesting for projects where the model must plan, execute, inspect evidence, recover from failures, and improve the artifact over time.

Qwen Team says the hosted model is available through Qwen Studio and its API. This page explains its fit for the Atoms product workflow; model selection and runtime availability can vary by workspace.