Alibaba Unveils Qwen3.7-Max AI Model and AI Factory Stack

Alibaba has introduced Qwen3.7-Max, a flagship artificial intelligence model engineered to handle complex, multi-step autonomous tasks [1]. The model is part of a broader "AI factory" strategy that includes new proprietary semiconductor hardware and server infrastructure designed to support long-horizon agentic workflows [2].

Key takeaways

• Qwen3.7-Max is capable of executing over 1,000 tool calls in a single autonomous run [1].
• The model features a 1-million-token context window and demonstrated a 10x improvement in inference speed compared to its predecessor [1, 2].
• Alibaba’s new hardware stack includes the Zhenwu M890 AI accelerator and the Panjiu AL128 server, which links 128 accelerators into a single unit [2].
• Independent assessments place Qwen3.7-Max as the highest-ranked Chinese model on the Artificial Analysis Intelligence Index, though it trails leading Western models like GPT-5.5 [2].

Autonomous Performance and Hardware Integration

The launch of Qwen3.7-Max marks a significant shift toward autonomous agentic capabilities, with the model demonstrating the ability to perform over 1,100 tool calls during a 35-hour autonomous run [1, 2]. During this demonstration, the model was tasked with writing and optimizing performance-critical software for Alibaba’s own Zhenwu M890 chip without access to existing documentation [2]. Alibaba reports that this process resulted in a 10x speedup on the model's kernel implementation [2]. While these results are first-party claims from Alibaba, the model has also performed well on public benchmarks, scoring 92.4 on the GPQA Diamond reasoning test and 91.6 on LiveCodeBench [1].

The model is supported by a vertically integrated hardware stack, including the Zhenwu M890 accelerator and the ICN Switch 1.0, which enables high-bandwidth communication between chips [2]. Although analysts note that the M890’s raw specifications trail leading Western chips like the Nvidia H200, the hardware is positioned as a viable domestic alternative for the Chinese market [2]. Alibaba has committed approximately $53 billion to its cloud and AI infrastructure over three years, signaling a long-term strategy to control both the model architecture and the underlying silicon [2].

Why it matters

Alibaba’s strategy reflects a push for "full-stack" sovereignty, allowing the company to tune its AI models specifically to its proprietary chip architecture [2]. This approach aims to solve the challenges of long-horizon, tool-heavy tasks that currently strain most AI systems [1]. While the model shows strong performance in coding and reasoning, its factual recall behavior has shifted; it now exhibits a lower hallucination rate, partly by choosing to abstain from answering questions rather than providing potentially incorrect information [2]. As Alibaba continues to integrate these tools into its cloud ecosystem, the industry will be watching to see how the model performs in real-world enterprise environments compared to established frontier models from Western labs [1, 2].