It’s the latest reasoning model designed to rival OpenAI’s o1-series, and here’s the kicker—it’s available for download under a permissive license.
That’s rare in this space.
Developed by Alibaba’s Qwen team, QwQ-32B-Preview packs a punch with 32.5 billion parameters and can handle prompts of up to 32,000 words. It has already outperformed OpenAI’s o1-preview and o1-mini on benchmarks like AIME and MATH.
The model shows prowess in solving logic puzzles and complex math problems, but Alibaba’s own testing reveals some quirks: it might suddenly switch languages or underperform on tasks needing common sense.
Oh, and did I mention? Unlike many other models, it’s built to fact-check itself, which reduces errors but takes longer to deliver results.
However, as a product of a Chinese company, its responses align with government regulations, dodging sensitive topics.
What does this mean for the AI space?
More open competition. More innovation. And possibly, a faster route to better AI.