Alibaba Cloud revamps Qwen platform to bring agents into production
At the 2026 Apsara Conference in Hangzhou on Sept. 22, Alibaba Group strategic vice president Wen Zheng announced a major upgrade to the Qwen AI platform, Alibaba Cloud's model service platform. The upgrade is centered on bringing agents into production, adding agent services and industry AI solutions on top of model services. New capabilities include API Express mode, Agent Studio, and Qwen AI cockpit solutions.
Wen said AI is still in an early stage, with headroom in user coverage, scenario penetration and compute scale. Over the past year, the customer count on Alibaba Cloud's MaaS platform grew sixfold. As business processes begin to initiate calls autonomously, agents are entering core enterprise workflows, and companies need a complete system covering model supply, production operations, agent construction and hosting.
To handle agent workloads that involve long decision chains, high concurrency and latency sensitivity, the Qwen platform uses FlashBoot, UniScheduler and other capabilities to orchestrate heterogeneous compute. Model cold-start time was cut from 1,200 seconds to 70 seconds, allowing 10,000 pods to be launched within a minute. First-token latency was reduced by 38%. The new API Express mode boosts TPS by 1.5 to 2 times.
Wen also introduced Agent Studio, an enterprise-grade full-stack agent platform that automatically routes tasks to suitable models, provides 24/7 hosted operations, and supports integration with internal knowledge and external software services. It offers more than 50 atomic APIs and a planned self-evolution engine. One Key MCP connects more than 100 ecosystem services with a single API key.
In industry applications, Fanhai Group built an industrial AI platform and CNC process agents on Qwen; Honor uses Qwen Intelligence to power its YOYO agent with multimodal understanding and long-chain task planning; and BYD developed a cockpit super-agent based on the Qwen AI cockpit solution. Wen said Alibaba will continue investing in full-stack AI, with Alibaba Cloud's compute scale expected to exceed 20GW by 2032.
