Alibaba Unveils QwQ-32B, a Compact Reasoning Model Rivaling DeepSeek-R1
On March 6, Alibaba released and open-sourced its new reasoning model, QwQ-32B, featuring 32 billion parameters. Despite being significantly smaller than DeepSeek-R1, which has 6,710 billion parameters (with 3.7 billion active), QwQ-32B matches its performance in various benchmarks. QwQ-32B excelled in math and coding tests, outperforming OpenAI’s o1-mini and distilled versions of DeepSeek-R1. It also […]
On March 6, Alibaba released and open-sourced its new reasoning model, QwQ-32B, featuring 32 billion parameters. Despite being significantly smaller than DeepSeek-R1, which has 6,710 billion parameters (with 3.7 billion active), QwQ-32B matches its performance in various benchmarks. QwQ-32B excelled in math and coding tests, outperforming OpenAI’s o1-mini and distilled versions of DeepSeek-R1. It also scored higher than DeepSeek-R1 in some evaluations like LiveBench and IFEval. The model leverages reinforcement learning and integrates agent capabilities for critical thinking and adaptive reasoning. Notably, QwQ-32B requires much less computational power, making it deployable on consumer-grade hardware. This release aligns with Alibaba’s AI strategy, which includes significant investments in cloud and AI infrastructure. Following the release, Alibaba’s US stock rose 8.61% to $141.03, with Hong Kong shares up over 7%.[Jiemian, in Chinese]
Get AI news in your inbox
Daily digest of what matters in AI.
Key Terms Explained
The AI company behind ChatGPT, GPT-4, DALL-E, and Whisper.
The ability of AI models to draw conclusions, solve problems logically, and work through multi-step challenges.
A learning approach where an agent learns by interacting with an environment and receiving rewards or penalties.