Alibaba Cloud Qwen — Alibaba's Cloud Computing Arm

By March 2026, open-source Qwen models exceeded 942 million downloads on Hugging Face — double the total downloads of the next eight competitors combined. The irony: the company behind this number is Alibaba, primarily known for e-commerce, not technical research.

Alibaba and the Shift Toward Artificial Intelligence

Jack Ma founded Alibaba in 1999 in Hangzhou with the aim of creating a digital infrastructure for commerce in China. As the company expanded into cloud services through Alibaba Cloud (Aliyun) in 2009, the massive computing capabilities that would later become the backbone of AI projects were formed.

In 2023, Tongyi Qianwen — later known as Qwen — launched as a closed beta in April, then as a public release in September after obtaining Chinese regulatory approvals. The core architecture was inspired by Llama’s structure but with substantial modifications and different training data.

The Qwen Model Family: Breadth and Diversity

What distinguishes Qwen from many competitors is the breadth and diversity of its model family:

Qwen 2.5, released in September 2024, came in variants ranging from 0.5 billion to 1.5 billion, 7 billion, 14 billion, 32 billion, and 72 billion parameters. This gradient enables deployment on diverse devices from smartphones to cloud data centers. Qwen 2.5 Coder was trained on 5.5 trillion tokens and supports 92 programming languages. Qwen 2.5 Math achieves 84% accuracy in solving mathematics problems.

Qwen 3, released in April 2025, was trained on 36 trillion tokens across 119 languages and dialects. It includes dense models from 0.6 to 32 billion parameters, and sparse MoE (Mixture of Experts) models from 30 billion (3 billion active) to 235 billion (22 billion active). The Qwen3-235B model competes with DeepSeek-R1, o1, and Grok 3 according to independent benchmarks.

Qwen 3.5, released in February 2026 with 397 billion parameters, supports 201 languages and dialects — doubling from 82 in the previous generation — with enhanced agent capabilities and compatibility with OpenClaw agents.

Specialized models complete the ecosystem: Qwen2.5-VL for vision, Qwen2.5-Audio for audio, and Qwen2.5-Omni, a multimodal model that accepts text, images, video, and audio, and outputs text and audio.

Open-Source Strategy and Its Global Impact

Alibaba’s decision to release most Qwen models under the open-source Apache 2.0 license was a strategic bet. The logic: when developers worldwide adopt Qwen as a foundation, they create projects and services that depend on its continuity and quality of performance, expanding influence even when Alibaba doesn’t directly collect payment for every download.

By March 2026, the number of models derived from Qwen exceeded 200,000 on Hugging Face. Airbnb announced that its customer service bot partly runs on Qwen models. This kind of adoption doesn’t generate direct revenue for Alibaba, but it solidifies presence.

Alibaba Cloud and the Enterprise Ecosystem

Qwen services are offered through Alibaba Cloud (Aliyun) under a competitive pricing model. Alibaba reduced inference prices by 97% in January 2026 amid a price war that ignited after DeepSeek’s emergence.

Qwen Studio combines chat functions, image and video understanding, image generation, document processing, and web search in a single platform. DingTalk — the enterprise collaboration platform — integrates Qwen into the workflows of 2.2 million organizations.

In March 2026, Alibaba consolidated all its AI brands under the single Qwen name, establishing “Tongyi Lab” as a centralized research laboratory.

Numbers and Market Position

Alibaba’s market value hovers around $400 billion (ticker BABA) after a series of internal structural changes and restructuring in 2023-2024. Total revenue for 2024 exceeded $130 billion.

Qwen App recorded 73.52 million daily active users during the 2026 Chinese Spring Festival holiday, a 940% increase — the fastest growth for any Chinese AI application during that period.

Challenges and Risks

In March 2026, Lin Junyang, the technical leader of the Qwen team, resigned, followed by multiple disruptions within the team. Alibaba shares in Hong Kong fell 5.3% on the day of the announcement, and senior management held an emergency meeting to contain the impact.

Geopolitical tensions complicate global expansion: government institutions and Western companies are reconsidering decisions to adopt Chinese models under increasing regulatory pressure.

2026-2027 Trajectory

After the technical leader’s departure, the pressing question is whether Alibaba will maintain the development momentum that achieved this leadership, or whether restructuring will threaten continuity. Qwen 3.5, released weeks before Lin’s departure, proved that the technical infrastructure is deeper than to stop with one person’s exit.

In the long term, the question remains: can an open-source model ecosystem be continuously funded by competitive cloud infrastructure, or does this model require clearer commercial gains than what broad free dissemination achieves?

---

Explore More

Want to learn more about the latest models mentioned here? Visit our Top AI Models List for a comprehensive comparison, or browse the Latest AI Tools to boost your productivity.