Why Alibaba Is Changing the Rules: Agents as the New Cloud Consumers

Alibaba Cloud CTO and President of International Business Dr. Li Feifei clearly named a trend that can no longer be ignored at the launch: "The demand for AI continues to surge. The explosion of agents in particular has triggered exponential growth in model calls and cloud resource consumption. When agents become the primary consumers of cloud services, every interface and interaction logic designed for humans must be rewritten."

In other words — cloud platforms like AWS, Google Cloud, or Azure were built on the assumption that a human would control them through a web interface or API. But the reality of 2026 is different: AI agents call APIs more often than humans do and need a completely different type of interface. Qwen Cloud is the first major platform that not only acknowledges this shift but builds its entire architecture around it.

Three Entry Points, One Platform: Skills, CLI, and Web

Qwen Cloud comes with a three-entry-point architecture, each optimized for a different type of user — whether it's a human, an automated workflow, or the AI agent itself:

Skills (for AI Agents)

The most compelling innovation is the concept of Skills Prompts — standardized natural language instructions that AI agents can directly read and execute. "Being agent-friendly is Qwen Cloud's core advantage," said Linlin Kong, head of Qwen Cloud. "We've wrapped complex capabilities like image generation, image understanding, or video processing into standardized instructions that agents can read."

In practice, this means the developer no longer has to search for endpoints, write API calls, test and debug, and then "teach" the agent how to work with them. A single natural language instruction is enough, and the agent itself understands what to do and carries it out immediately. No code under the hood — the agent learns and acts instantly.

CLI (for Workflows)

For automated deployment, batch job processing, and CI/CD pipelines, Qwen Cloud offers a high-speed command-line interface. The CLI layer is deterministic and reproducible — ideal for environments where you need stable and predictable results without depending on a graphical interface.

Web (for Humans)

For developers who want to explore the platform, a web portal is ready. It allows you to compare hundreds of models side by side, test them online, and get an API key with a single click. The entire journey from experiment to production happens in one browser window — with no need to tie up a credit card.

Six Modalities on a Single Platform

Most AI platforms start and end with text. Qwen Cloud goes significantly further. It combines Alibaba's complete model family with third-party models and covers six modalities:

• Text — conversation, summarization, translation
• Vision — document and screenshot analysis
• Image — image generation
• Video — video processing and summarization
• Speech — voice interaction
• Embedding — vector representations for search

For developers, this means one crucial thing: one API key, one integration — and you can do everything. As the application grows, the platform scales with it. No migration between different providers for different modalities.

DeepSeek and an Open Ecosystem

Qwen Cloud doesn't bet solely on its own models. The platform has already integrated DeepSeek, which is optimized for high-performance reasoning, and more models are continuously being added. Developers can freely switch between models without changing their integration code and intuitively compare cost and performance.

This open approach is also beneficial for Czech companies and developers — you don't have to lock yourself into a single provider and can choose the model that best fits your use case and budget.

Pricing and Enterprise Features: Token Plan Brings Predictability

One of the biggest pain points for companies adopting AI is uncontrollable costs. Qwen Cloud addresses this with several tools:

Token Plan — a multimodal subscription with a fixed monthly price that covers text, image, and speech. It solves the budget volatility of pay-as-you-go models. It offers shared seats for team collaboration and flexible quota overrun — if you exhaust the limit, the service isn't interrupted, and excess consumption is automatically converted to pay-as-you-go.

For larger organizations, the platform offers workspace isolation, detailed permission management, rate limiting, API key management, and transparent reporting — managers see exactly who consumed what, on which model, and through which channel. Cost analysis runs by API Key ID, Workspace ID, model name, input/output type, and call channel.

Specific Token Plan pricing hasn't been published yet, but Alibaba promises that the basic tier will remain free for experimentation — with no need to enter a payment card.

Availability in the Czech Republic and Europe

Qwen Cloud launches in Singapore as a global product. For European and Czech developers, this means the platform is available immediately through the web interface at qwen.ai. API calls from Europe will be routed through Singapore infrastructure, which may mean slightly higher latency compared to platforms with European data centers.

From the perspective of the EU AI Act, it will be key how Alibaba Cloud sets up data processing — whether it will offer European servers for companies that need to keep data within the EU. This has not yet been clarified.

Czech as a language isn't explicitly supported at the platform level, but individual models (especially Qwen 3.7) have very good Czech language support — both for understanding and generating text.

Developer Hackathon: The Challenge to Build a Production Agent

At launch, Alibaba announced a hackathon for developers who want to build production-ready AI agents on Qwen Cloud. This is a clear signal that the platform wants to build a community from the start and demonstrate what an agent-designed architecture can do in real-world scenarios.

Linlin Kong summed it up in three steps: "Open an account. Try a model. Five minutes. Give your agent a Skills prompt and CLI. One instruction, the whole platform. Get an API key — it works with everything you already use. The fastest path from a powerful model to real work."