Alibaba Launches Qwen3.7-Max for Long-Horizon Autonomous Agent Tasks

May 21, 2026

Alibaba, the group behind the Qwen series, launched Qwen3.7-Max as a flagship foundation model engineered for autonomous agents. It introduces long-horizon autonomy, maintaining execution loops for over 35 hours. This release follows the arrival of Qwen3.5-Plus on Vercel AI Gateway.

Autonomy duration: 35 hours
Tool calls: 1,000+ in single task
Performance gain: 10.0x over Triton reference
Integrations: MCP, multi-agent orchestration
Availability: Alibaba Model Studio API, Qwen Studio

The update addresses the reliability gap in multi-step reasoning where models often fail during extended tool-use sequences. By completing over 1,000 tool calls in a single session, Qwen3.7-Max mirrors the focus on persistent reasoning seen in GPT-5.5's agentic workflows. It is also available through OpenRouter's Qwen3.7-Max integration.

You can deploy the model for end-to-end coding across frameworks like Claude Code. Its native support for the Model Context Protocol enables integration with office productivity tools. This autonomy makes it suitable for high-reasoning tasks similar to those performed by Vercel's deepsec security agents. Access is available via Alibaba Model Studio.

View the full update on qwen.ai

Qwen

@Alibaba_Qwen8h ago

📣Meet Qwen3.7-Max — our latest flagship, made for the Agent Era. A versatile foundation for agents that actually get things done: 🧑‍💻 Coding agent, end to end. Frontend prototypes, multi-file refactors, real debugging — nails it. 🗂️ A reliable office and productivity assistant. Get your work done through MCP integrations and multi-agent orchestration. ⏱️ Long-horizon autonomy. 35 hours straight on a kernel optimization task — 1,000+ tool calls, zero hand-holding. 🔌 Scaffold-agnostic. Claude Code, OpenClaw, Qwen Code, or your own stack. Consistent reliability everywhere. API's up on Alibaba Model Studio. You can also take it for a spin on Qwen Studio. Go build something wild!🏃🏃‍♂️ 📖 Blog: https://t.co/y3AupX3Pa0 ✅ Qwen Studio: https://t.co/qpTnrCBjWt ⚡️ API：https://t.co/0sys00osKn

3162.4k

View on X

Still wondering? A few quick answers below.

Qwen3.7-Max is the latest flagship foundation model from Alibaba, specifically designed for the agent era. It serves as a versatile base for autonomous AI agents that can plan and execute complex, multi-step tasks independently. The model is optimized for long-horizon autonomy, meaning it can maintain continuous operation on difficult engineering or productivity tasks for extended periods.

The model is capable of end-to-end agentic coding, which includes creating frontend prototypes, performing multi-file refactors, and conducting real-world debugging. In a demonstration of its capabilities, it spent 35 hours autonomously optimizing a kernel, performing over 1,000 tool calls and 432 evaluations to achieve a tenfold speedup over standard reference benchmarks without any human intervention.

Qwen3.7-Max is designed to be scaffold-agnostic, meaning it works reliably across various agent frameworks such as Claude Code, OpenClaw, and Alibaba's own Qwen Code. It also features native support for the Model Context Protocol, an open standard that allows the model to connect easily to external data sources and office productivity tools for multi-agent orchestration.

Developers can access the model through the API available on Alibaba Model Studio. For those who want to test its capabilities before integrating it into their own applications, the model is also available for interactive use on Qwen Studio. It is intended to be used as a reliable foundation for building custom autonomous agents and productivity assistants.

Unlike standard conversational models, Qwen3.7-Max is specifically tuned for long-horizon autonomy and complex tool use. It is built to handle the iterative cycle of observing an environment, reasoning about the next step, and executing actions over thousands of turns. This focus on agentic engineering allows it to solve deep technical problems that require persistent, autonomous execution.