Alibaba enables broad access to its AI technology and innovations by releasing cutting-edge large models from the Qwen family as open source and hosting China’s largest open-source community platform, ModelScope.

Alibaba champions open source because the mission is to level the playing field for the little guys (small businesses, entrepreneurs, anyone with an idea) by helping them leverage innovation and technology to grow and compete more effectively in the economy. Through commitment to open source, it aims to promote:

• Democratization of AI: Open source reduces the barrier to entry, allowing resource-constrained startups, small businesses, researchers, and entrepreneurial developers to access cutting-edge technology and build their own models more cost-effectively.

• Advancing Scientific Research: Open source accelerates scientific breakthroughs by enabling researchers to experiment, validate findings, and contribute improvements.
  
• Transparency and Trust: Open source allows researchers and developers to inspect code and training data, reducing the “black box” nature of close models.

• Innovation and Collaboration: Open source promotes innovation by enabling the global developer community to improve and build on existing models. It also fosters a more competitive and diverse AI ecosystem.

Alibaba was one of the earliest major global tech companies to open source its self-developed large-scale AI model. It released its first open-model Qwen (Qwen-7B) in August 2023, with subsequent iterations encompassing language, multimodal, mathematical, and code models. The Qwen family open models have repeatedly topped the HuggingFace Open LLM Leaderboards, with performances matching leading global AI models across recognized benchmarks.

Alibaba was also the first company in the world to release a multi-modal reasoning AI model, QwQ (Qwen with Questions), as open-source.

Last month, Alibaba released our next-generation multimodal model, Qwen2.5-VL, and the latest large language model, Qwen2.5-1M, in open-source format. Both deliver top results across globally recognized benchmarks.

As of February 2025, over 90,000 Qwen-based derivative models have been developed in the global open-source community HuggingFace. This surpasses the number of Meta’s Llama-based derivative models and reflects Qwen’s popularity as one of the world’s most widely adopted open-source models.

Timeline of Alibaba’s Open-Source AI Initiatives

2022

November: Alibaba launched ModelScope, a platform that provides access to open-source AI models, including computer vision, natural language processing, and audio processing. These ready-to-deploy models were pre-trained to excel at tasks such as text-to-image generation.

2023

August: Alibaba published its first open-source model, the 7-billion-parameter LLM Qwen-7B and Qwen-7B-Chat, through ModelScope and Hugging Face. The models’ code, model weights, and documentation were made freely accessible to academics, researchers, and commercial institutions worldwide.

Late August: Alibaba launched two open-source large vision language models (LVLM), Qwen-VL and its conversationally fine-tuned Qwen-VL-Chat. Qwen-VL is the multi-modal version of Qwen-7B. Capable of understanding both image inputs and text prompts in English and Chinese, Qwen-VL can perform various tasks, such as responding to open-ended queries related to different images and generating image captions.

December: Alibaba released the 72-billion-parameter and 1.8-billion-parameter versions of Qwen, Qwen-72 B, and Qwen-1.8B as open models. It also made more multi-modal LLMs freely available, including Qwen-Audio and Qwen-Audio-Chat, a pre-trained audio understanding model and its conversationally fine-tuned version for research and commercial purposes.

2024

June: Alibaba released the open-model Qwen2 series, which encompasses several base language models and instruction-tuned language models. Qwen2-72B model outperformed other leading global open-source models on 15 benchmarks, including language understanding, language generation, multilingual capability, coding, mathematics, and reasoning.

September: Alibaba unveiled 100 new open-source Qwen2.5 models ranging from 0.5 to 72 billion parameters in size, feature enhanced knowledge and more substantial capabilities in math and coding, and can support over 29 languages.

November: Alibaba released a reasoning AI model QwQ (Qwen with Questions). QwQ-32B-Preview, an open-source experimental research model with 32 billion parameters, showcased impressive analytical capabilities. QwQ was the first reasoning AI model in the world made available as open-source.

December: Alibaba introduced QVQ-72B-Preview (“QVQ”), an open-sourced, experimental research model designed to advance visual reasoning capabilities.

2025

January: Alibaba released its latest visual-language open model, Qwen2.5-VL. The model demonstrates remarkable multimodal capabilities and serves as a visual agent to facilitate the execution of simple tasks on computers and mobile devices by directing the use of various tools.

Late January: Alibaba unveiled the newest version of the Qwen large language model, Qwen2.5-1M. This open-source iteration is distinguished by its capability to process long context inputs and handle up to 1 million tokens.

Late February: Alibaba Cloud open-sources 14 bln and 1.3 bln parameter versions of the Wan2.1 series, the latest iteration of its video foundation model Tongyi Wanxiang (Wan).