Alibaba Qwen3.7-Max runs autonomously 35 hours

Alibaba Cloud said on May 21 that its new proprietary model, Qwen3.7-Max, is built for long-running agent tasks rather than short chat turns. In a company post, Alibaba said the model sustained a 35-hour autonomous kernel-optimization run with more than 1,000 tool calls. (alibabacloud.com) VentureBeat reported the same day that Alibaba was also positioning the model as interoperable with external agent harnesses, including Anthropic’s Claude Code. Alibaba’s description matters because the company is not only claiming raw model capability. It is also claiming that Qwen3.7-Max can plug into software that developers already use to run coding and workflow agents. (alibabacloud.com) Alibaba said the model “generalizes across agent scaffolds” including Claude Code, OpenClaw and Qwen Code, and supports office workflows through MCP integrations and multi-agent orchestration. ### What exactly did Alibaba say the model did for 35 hours? Alibaba said Qwen3.7-Max completed a “35-hour, fully autonomous kernel optimization run” comprising more than 1,000 tool calls. The company described that test as evidence that the model can sustain coherent reasoning over long horizons and across hundreds or thousands of steps. (alibabacloud.com) The 35-hour figure is a claim about endurance, not just single-response quality. Alibaba’s post tied the run to coding and engineering work, saying the model is meant for tasks ranging from frontend prototyping to complex multi-file software engineering. (alibabacloud.com) ### How is Claude Code part of this story if Qwen is Alibaba’s model? Alibaba said Qwen3.7-Max performs consistently when deployed through Claude Code, OpenClaw, Qwen Code and other frameworks. That means Alibaba is presenting the model as the intelligence layer inside external agent tooling rather than requiring developers to use only Alibaba’s own interface. (alibabacloud.com) Anthropic’s GitHub repository describes Claude Code as an “agentic coding tool” that runs in the terminal and can execute routine coding tasks, explain code and handle git workflows through natural-language commands. Alibaba’s compatibility claim therefore points to a setup in which developers keep the Claude Code workflow but swap in a Qwen model behind the API layer. (alibabacloud.com) ### What does “Anthropic API-compatible” mean here? Alibaba Cloud’s documentation says Claude Code works with Qwen models through Model Studio’s “Anthropic API-compatible interface.” The setup instructions tell users to point `ANTHROPIC_BASE_URL` to Alibaba’s endpoint and use a Model Studio API key, while keeping the Claude Code client in place. (alibabacloud.com) Alibaba’s published example currently lists Qwen-Max, Qwen-Plus, Qwen-Flash and other model families in that compatible service. The company’s May 21 post separately said Qwen3.7-Max would be “available soon” via Model Studio. (github.com) ### Is this an Alibaba-only framework or a broader interoperability push? Alibaba said Qwen3.7-Max is designed as a “versatile agent foundation” for coding, office automation and long-horizon execution. The company also referenced MCP integrations, multi-agent orchestration and benchmark setups that use external harnesses such as Claude-code and Harbor/Terminus-2. (alibabacloud.com) VentureBeat’s May 21 report highlighted the same point from a developer-adoption angle: Alibaba is pitching Qwen3.7-Max as a model that can work inside existing agent ecosystems rather than one locked to a proprietary shell. (alibabacloud.com) ### Where can developers actually use it next? Alibaba said on May 21 that Qwen3.7-Max would be available soon through Alibaba Cloud Model Studio. Alibaba Cloud’s Claude Code documentation was already live as of March 3 and applies to the company’s international service in the Singapore region, with environment-variable setup instructions for developers who want to connect Claude Code to Qwen models. (alibabacloud.com) (venturebeat.com)