The future of model service platforms isn’t human — it’s agents. Today, at its summit, Alibaba Cloud unveiled Qianwen Cloud (www.qianwenai.com), the first AI product website built specifically for agents. It provides API access to over 150 mainstream models, but the real story is how it’s been rearchitected: the entire model service chain has been refactored into Skills, MCP (Model Context Protocol), and CLI interfaces, allowing agents to directly understand and autonomously invoke every capability on the site.
Unlike traditional console interfaces designed for human developers, Qianwen Cloud ditches the tedious process of reading documentation, hand-writing integration code, and debugging API parameters. The reason is simple: traditional APIs are built for human clicks and manual integration; agents need lightning-fast transient calls. With this new architecture, agents like OpenClaw, Hermes Agent, and Claude Code can learn all core features with a single command.
On the technical side, Qianwen Cloud Skills package the full pipeline of model selection, invocation, authentication, and usage querying. The agent can dynamically route based on the task — automatically calling a vision model for images or switching to an image generation model for illustrations, enabling automated multi-model orchestration. The Qianwen Cloud CLI client covers login, status queries, and environment diagnostics, letting agents run entire workflows via command line or scripts in the background. Both toolkits are open-sourced on GitHub.
Beyond interface restructuring, Qianwen Cloud also lets agents dive into cost management. Agents can pull real-time high-frequency usage data, analyze consumption trends, identify anomalies, and autonomously offer cost optimization suggestions. To handle new usage patterns like high-frequency AI coding and multi-agent collaboration, Qianwen Cloud introduces a Token Plan subscription model alongside standard pay-as-you-go pricing, aimed at significantly reducing token costs in agent scenarios.
