Alibaba has released Qwen with Questions (QwQ), a new artificial intelligence reasoning model designed to compete with OpenAI’s o1 system. The model features 32 billion parameters and can process contexts of up to 32,000 tokens.
According to Alibaba’s testing, QwQ outperforms OpenAI’s o1-preview on mathematical and scientific reasoning benchmarks AIME and MATH. The company states that QwQ demonstrates superior performance in mathematical problem-solving, though it shows lower results than o1 in coding tasks.
QwQ employs additional computing cycles during inference to review and correct its responses, a technique known as inference-time scaling. This approach allows the model to improve its performance on tasks requiring logical reasoning and planning.
The model demonstrates some limitations, including unexpected language switching and occasional circular reasoning loops. These limitations have been openly acknowledged by Alibaba in their release documentation.
The model is released under an Apache 2.0 license, making it available for commercial use. However, Alibaba has not published a detailed paper describing the training data and processes, limiting the ability to reproduce the model’s results.
The release comes amid industry discussions about the diminishing returns of traditional model scaling approaches. Major AI laboratories, including OpenAI, Google DeepMind, and Anthropic, are reportedly experiencing challenges in improving model performance through increased size alone.
Sources: VentureBeat, TechCrunch, VentureBeat