Alibaba next to outperform OpenAI’s o1?

The Chinese tech firm, Alibaba, has released a new AI model, called QwQ-32B-Preview (or QWQ for short), which has advanced reasoning capabilities, meaning it can ‘reason’ its way through tasks and answer complex multi-step questions on difficult subjects like logic, Math, and coding. This release puts QWQ in direct competition with OpenAI’s recently released AI model, o1-preview.

According to benchmark test results, QWQ scored higher than OpenAI’s o1 (and o1-mini) in Math benchmarks (which assess reasoning and mathematical capabilities), achieving an accuracy rate of 90.6% compared to o1’s 85.5%. It also scored higher in Aime tests (which use other AI models to test performance), scoring 50% vs o1’s 44.6% (for added context, OpenAI’s chatbot GPT-4 only scored 9.3% in these tests).

Although QWQ doesn’t accept more than 32,000-word prompts (o1 accepts up to 96,000), refuses to answer certain political questions (due to pressure from the Chinese government to build AI models that embody core socialist values”), and is prone to switch languages unexpectedly, get stuck in loops, and underperform on “common sense reasoning” tasks, it has been released as an open-source model. This means it can be downloaded (from the dev platform, Hugging Face) and used for commercial purposes. However, only certain parts of the model have been released, so users can’t replicate the model or see its inner workings.