A new “reasoning” AI model, QwQ-32B-Preview, developed by Alibaba’s Qwen team, is challenging OpenAI’s models in performance benchmarks. With 32.5 billion parameters, this model can solve logic puzzles and challenging math questions, although it may struggle with common sense reasoning and language switches. QwQ-32B-Preview can fact-check itself and avoid pitfalls, but this also means it takes longer to solve problems. Similar to OpenAI’s approach, this model reasons through tasks and plans ahead for its actions. Unlike most AI systems, QwQ-32B-Preview is openly available under an Apache 2.0 license and has been benchmarked in China to align with socialist values. The model has specific components released, making it challenging to replicate or understand fully. The rise of reasoning models highlights a shift towards new development techniques like test-time compute, with major AI labs and Chinese firms leading the way in this area.
https://techcrunch.com/2024/11/27/alibaba-releases-an-open-challenger-to-openais-o1-reasoning-model/