Qwen and Wan models to be open source according to modelscope

Alibaba Cloud has announced a major strategic shift by making its flagship Qwen and Wan series of large language models (LLMs) fully open-source. The models, including the high-performance Qwen2.5-72B, are being released on Alibaba's AI model community platform, ModelScope. This move directly challenges the closed, proprietary approach of leading AI labs like OpenAI and Google, offering the global developer community free access to state-of-the-art AI capabilities that were previously gated behind APIs or commercial licenses.

By open-sourcing Qwen and Wan, Alibaba is providing not just model weights but also the associated toolchains, documentation, and commercial usage rights. This enables developers to run, fine-tune, and deploy these models on their own infrastructure without restrictive licensing fees. The Qwen2.5 series, in particular, has demonstrated benchmark performance competitive with models like GPT-4 and Claude 3, making this a significant contribution to the open-source AI landscape. This release accelerates innovation by allowing startups and researchers to build upon a powerful base model without the massive computational cost of training from scratch.

The decision reflects a growing trend where major tech companies leverage open-source to build ecosystem influence and developer mindshare. For Alibaba, it's a play to establish ModelScope as a central hub for AI development in China and beyond, competing with platforms like Hugging Face. The availability of such advanced models as open-source could pressure other tech giants to follow suit, potentially democratizing access to cutting-edge AI and reducing industry reliance on a handful of proprietary API providers.