Alibaba Unveils "Qwen3.7-Max" — an Agent-Focused Model Capable of 35 Hours of Autonomous Execution, Now Closed
On May 19, 2026, Alibaba unveiled its new flagship model, “Qwen3.7-Max.” Ahead of the announcement at the Alibaba Cloud Summit, API access went live the same day on Alibaba Cloud Model Studio. The model was designed for the “agent era” — built for coding, automating office work, and long-running tasks spanning hundreds or thousands of steps.
Details
- Context length: Equipped with a 1 million-token context window, a major expansion from the 256,000 tokens of the previous version, Qwen3.6 Max Preview
- Long-horizon autonomous execution demo: In a kernel optimization demo, the model reportedly executed more than 1,000 tool calls and code fixes fully autonomously over 35 hours, improving inference speed roughly 10x over the previous version
- Intended use cases: Designed primarily for coding and debugging, automating office workflows, and carrying out long-running agentic tasks
- Pricing: $2.50 per million input tokens, $7.50 per million output tokens, and $0.25 for cached input
- Availability: A closed model with undisclosed weights — not published on Hugging Face or elsewhere. Available only via API through Alibaba Cloud Model Studio
What happened next
The launch of Qwen3.7-Max underscored a shift in Alibaba’s strategy for its top-tier models, moving away from open weights toward an API-only, closed approach. A clear “two-tier” strategy has emerged, in which open-weight releases are now handled mainly by mid-tier models such as Qwen3.7-Plus.