Alibaba Releases Qwen3.7-Max AI Model for Autonomous Tasks

Alibaba’s Qwen team has introduced Qwen3.7-Max, a proprietary AI model designed to perform complex, multi-day autonomous tasks [1]. The model is capable of executing over 1,000 tool calls in a single session and is built to handle long-horizon reasoning, allowing it to maintain focus on intricate projects without the performance degradation often seen in earlier language models [1, 2].

Key takeaways

• Qwen3.7-Max demonstrated the ability to operate autonomously for 35 continuous hours, completing 1,158 tool calls during a kernel optimization task [1].
• The model supports cross-harness generalization, allowing it to integrate with frameworks like Claude Code and OpenClaw [1, 2].
• Alibaba reports the model achieved a 92.4 score on the GPQA Diamond benchmark and 80.4 on the SWE-Verified software engineering test [2].
• Access to the model is provided through Alibaba Cloud Model Studio, with pricing set at $2.50 per million input tokens and $7.50 per million output tokens [1].

Engineering and Agentic Capabilities

The development of Qwen3.7-Max focuses on "environment scaling," where the model is trained across diverse agentic environments to improve its ability to navigate multi-step decision-making [1]. In one demonstration, the model autonomously diagnosed and corrected compilation failures while optimizing an attention kernel, achieving a 10.0x geometric mean speedup [1]. This performance surpassed other Chinese models, such as GLM-5.1 and Kimi K2.6, which Alibaba noted often terminated sessions when progress stalled [1].

Beyond technical coding tasks, the model is designed to function as a "versatile agent foundation" for enterprise automation [1]. It features a 1-million-token context window and has been tested in simulations such as "YC-Bench," where it managed a startup's one-year lifecycle, including personnel management and contract screening [1]. The model also includes self-monitoring features to detect and correct instances of reward-hacking during training [1].

Strategic Positioning in the AI Market

Alibaba has opted to release Qwen3.7-Max as a proprietary model rather than an open-source tool, aligning its strategy with Western firms like Google and OpenAI [1]. By offering the model through paid APIs, the company aims to recoup the significant costs associated with training high-performance AI [1]. While the model is available for global developers via Alibaba Cloud, its reliance on Chinese-based endpoints may present compliance or data sovereignty challenges for some Western enterprises [1].

The model is positioned to compete with top-tier global AI labs, with benchmark scores on LiveCodeBench and GPQA Diamond placing it among the most capable models currently available [2]. As the industry shifts toward the "agent era," where AI models are expected to plan and execute tasks over extended periods, Qwen3.7-Max represents Alibaba’s effort to provide a cognitive engine for software development pipelines and complex financial reporting [1, 2].