Alibaba Unveils Qwen QwQ-32B: A Compact Open‑Source LLM Rivaling DeepSeek

Since DeepSeek sparked global interest, Chinese open‑source large models have surged, and Alibaba now introduces its strongest competitor: the Qwen QwQ‑32B model.

QwQ‑32B, with only 320 billion parameters, delivers performance comparable to DeepSeek‑R1’s 6,710 billion‑parameter model, thanks to large‑scale reinforcement learning that improves mathematical, coding, and general reasoning abilities.

The model is released under a permissive Apache 2.0 license, allowing free download and commercial use worldwide.

It can be run locally on consumer‑grade graphics cards, dramatically lowering deployment costs, and is already accessible via the chat.qwen.ai website (QwQ‑32B‑Preview) and the Tongyi app.

Benchmark results show QwQ‑32B matches DeepSeek‑R1 on the AIME24 math set and LiveCodeBench coding tests, outperforming o1‑mini and similarly sized distilled models. It also exceeds DeepSeek‑R1 on the LiveBench, IFEval, and BFCL evaluations.

Alibaba highlights that the model’s success demonstrates how large‑scale reinforcement learning can boost foundational models toward general AI.

Alongside the technical announcement, the article promotes a paid “DeepSeek scenario practice collection” and a “AI Club” community, offering various AI tools, tutorials, and exclusive perks for a fee, with limited‑time discounts and incentives.

Finally, the post notes that Alibaba’s stock rose 6.7 % following the release, underscoring the market impact of the new model.