Alibaba goes open-source to take on Meta's Llama

Qwen-7B and Qwen-7B-Chat are two open-source, 7-billion-parameter language models from Alibaba, built to help developers create bilingual generative AI and conversational tools.

Both models stem from Alibaba's Tongyi Qianwen foundation model launched in April. Qwen-7B serves as the base model, while Qwen-7B-Chat is fine-tuned specifically for conversational applications.

What it can do

With these two 7B models, researchers, academics, and enterprise developers can build generative AI apps without the massive time and cost required to train base models from scratch.

Both models support text generation in English and Chinese. That gives teams a turnkey foundation for building cross-lingual chatbots, drafting tools, and custom workflow automations.

Alibaba has already deployed this tech stack internally, including rolling out Tongyi Wanxiang last month for text-to-image generation. This open-source release hands that same core technology to third-party builders.

Licensing and commercial terms

Global developers and academic researchers can access Qwen-7B and Qwen-7B-Chat for free.

Commercial deployment is permitted without fee, with one condition: companies with more than 100 million monthly active users must request a royalty-free license from Alibaba.

Alibaba is not collecting licensing revenue here. The release is purely about distribution and getting the models into as many developer workflows as possible.

Why it matters

If you need a mid-sized open model with native bilingual Chinese and English capabilities, Qwen-7B is worth an immediate look. For almost every startup or indie builder well below the 100-million active user cutoff, it is essentially frictionless.

Meta proved the power of the open-source playbook with Llama, and Microsoft even made Llama 2 available on its Azure cloud service last month. Alibaba is now bringing that exact fight to the bilingual domain.

We think the move is clear: give the weights away to capture developer mindshare, and let Alibaba Cloud reap the compute revenues once those AI applications take off.