Alibaba’s Open-Weight AI Models Hit 3 Billion Downloads in the past six months, surpassing Meta and Google, according to Hugging Face data reported by Bloomberg.

Alibaba’s open-weight models hit 3 billion downloads, surpassing Meta and Google

Alibaba’s open-weight models have accumulated 3B+ global downloads in the past six months, overtaking Meta and Google to become the world’s most downloaded AI model family. The milestone, reported by Bloomberg and backed by Hugging Face data, underscores the rapid growth of Alibaba’s Qwen ecosystem and the broader shift toward open-weight AI.

According to an emailed statement from Alibaba, its Qwen family of models has open-sourced more than 460 models, and the ecosystem has already spawned over 300,000 derivative models. The company says these downloads span developers, researchers, startups, and enterprises across multiple regions, including Southeast Asia and Africa, where Alibaba Cloud is actively promoting Qwen-based solutions.finance.yahoo+1

Hugging Face data: Alibaba vs Meta and Google

Hugging Face’s latest “State of Open Models” report, published on August 14, 2026, provides a clear comparison for 2026:

• Alibaba (Qwen): over 3 billion downloads in the past six months (cumulative across platforms, with a very large share on Hugging Face).
• Google: about 418 million downloads in 2026.
• Meta: about 227 million downloads in 2026.

One Chinese-language summary of the report notes that Qwen alone accounted for roughly 2.045 billion downloads on Hugging Face Hub in 2026, excluding other platforms such as Alibaba’s own ModelScope community, indicating that the total figure is even larger.

Why this matters

The download gap is significant because it reflects not just interest, but actual adoption. Open-weight models allow developers to:

• Download model parameters and run them locally or on private infrastructure.
• Fine-tune models for specific industries, languages, and use cases.
• Integrate AI into products without depending entirely on hosted APIs.

Alibaba’s strategy of releasing a wide range of model sizes — from small, device-friendly variants to large, frontier-class models — has made Qwen attractive for both experimentation and production deployment. The sheer number of derivative models suggests that Qwen is becoming a base layer for many downstream applications.finance.yahoo+1

Competitive context

For Meta and Google, the numbers highlight a changing landscape. Meta’s Llama family and Google’s open models have been influential, but Alibaba’s aggressive release cadence, permissive licensing, and strong performance across benchmarks appear to be resonating with the global developer community.

At the same time, high download volumes do not automatically translate into revenue. Alibaba will still need to convert this ecosystem growth into cloud usage, enterprise contracts, and paid services. However, owning one of the most widely adopted open model families gives Alibaba a powerful position in the AI value chain.

Summary: Alibaba’s open-weight Qwen models have surpassed 3 billion downloads in six months, far ahead of Google’s 418 million and Meta’s 227 million in 2026, according to Hugging Face data reported by Bloomberg. The milestone reflects rapid global adoption and a growing ecosystem of derivative models.