Alibaba's Qwen Tops Llama as Most Downloaded AI Model

Alibaba's Qwen Surpasses Meta's Llama as the Most Downloaded AI Model

Alibaba's Qwen AI has skyrocketed to prominence, overtaking Meta's Llama as the most downloaded AI model family with over 600 million global downloads and rapid app adoption exceeding 10 million in its first week alone. This milestone underscores China's rising dominance in open-source AI, challenging Western giants like OpenAI and Meta through free access, superior benchmarks, and seamless enterprise integration.

The Explosive Launch of the Qwen AI App

Alibaba's Qwen AI app, powered by its open-source model family, achieved more than 10 million downloads within seven days of its public beta release in mid-November, outpacing the early growth of ChatGPT, DeepSeek, and Sora. Unlike subscription-based competitors unavailable in mainland China, Qwen offers free access and integrates deeply with Alibaba's ecosystem, including e-commerce, mapping, and local business tools, enabling "agentic AI" for cross-scenario tasks beyond simple chat.

Designed as a comprehensive tool for professional and personal use, the app supports deep research, image generation, and slide creation, positioning Alibaba as a serious rival to Google's Gemini and ChatGPT. This consumer traction boosted Alibaba's Hong Kong-listed shares by 4.67% on the announcement day.

Open-Source Success: 600 Million Downloads and Beyond

Since open-sourcing the first Qwen generation in 2023, Alibaba has released over 300 AI models, amassing more than 600 million downloads and inspiring 170,000+ derivative models worldwide. On platforms like Hugging Face and GitHub, Qwen models alone exceed 20 million downloads, cementing its status as one of the most adopted open-source large language model series.

• Over 90,000 enterprises use Qwen via Alibaba Cloud's Model Studio, with 2.2 million corporate users accessing it through DingTalk.
• More than 1 million corporates and individuals have built on Qwen using Model Studio, creating over 800,000 agents for scenarios like content creation and smart home management.
• Model calls on Model Studio surged 15 times in the past 12 months, reflecting booming demand.

Technical Prowess: Qwen3-Max Leads Global Benchmarks

Qwen3-Max, Alibaba's largest model with over 1 trillion parameters, ranks among the top three globally, excelling in code generation (69.6 on SWE-Bench) and agentic tasks (Tau2-Bench). Trained on up to 20 trillion tokens with refinements like supervised fine-tuning and reinforcement learning, it handles language, coding, and multimodal processing.

High-profile endorsements highlight its edge: Airbnb CEO Brian Chesky revealed heavy reliance on Qwen, while NVIDIA CEO Jensen Huang noted its dominance in open-source AI. The Qwen 3 series, released April 2025 under Apache 2.0 license, supports 128K+ context lengths, 119 languages, and hybrid reasoning modes, outperforming rivals in STEM, math, and coding benchmarks like ArenaHard (95.6) and AIME25 (81.6).

Enterprise and Global Impact

Qwen's strategy blends consumer appeal with enterprise focus, disrupting subscription models by embedding AI into ecosystems like Alibaba Cloud. Its traction in Silicon Valley and recognition on Fortune's 2025 Change the World list affirm Alibaba's leadership in democratizing AI through open-source innovation.

Conclusion: A New Era for Open-Source AI

Alibaba's Qwen has not only eclipsed Meta's Llama in downloads but redefined AI accessibility with unmatched scale, performance, and adoption. As Qwen continues to evolve, it signals a competitive shift where open-source models from China are setting global standards.

This trajectory promises broader AI innovation, empowering developers, enterprises, and users worldwide with powerful, free tools for the future.