QwQ-32B-Preview is available under an Apache 2.0 license for commercial applications. However, only certain components of the model have been released, making it impossible to replicate or gain much insight into the system’s inner workings. The model's development comes amid scrutiny of "scaling laws" and a scramble for new AI approaches. Big labs, including Google, are reportedly betting on test-time compute, which underpins models like QwQ-32B-Preview, as the future of AI.
Key takeaways:
- A new AI model, QwQ-32B-Preview, developed by Alibaba's Qwen team, rivals OpenAI’s o1 and is the first to be available for download under a permissive license.
- QwQ-32B-Preview, which contains 32.5 billion parameters and can consider prompts up to 32,000 words in length, performs better on certain benchmarks than OpenAI's models.
- Despite its advanced reasoning capabilities, QwQ-32B-Preview has limitations such as switching languages unexpectedly, getting stuck in loops, and underperforming on tasks that require common sense reasoning.
- The model is available under an Apache 2.0 license for commercial applications, but only certain components have been released, making it impossible to fully replicate or gain insight into the system’s inner workings.