Alibaba Releasing An ‘Open’ Challenger To OpenAI’ With o1 Reasoning Model

Avijit SahNovember 28, 2024

1 7 2 minutes read

Alibaba Releasing An Open Challenger To OpenAI

QwQ-32B preview holds more than 32.5 billion parameters and it was developed by Alibaba’s Qwen team. It can consider the prompts up to ~32,000 words in length. It also performs better on certain benchmarks than the o1-preview and o1-mini. In these two resonating models that OpenAI released so far the parameters roughly correspond to the model’s problem-solving skills. The models with more parameters are generally performing better than the fewer parameters. Open AI does not actually disclose the parameter counts for its model.

According to Alibaba’s testing session, QwQ-32B-Preview beats OpenAI’s o1 models in the AIME and MATH trials. AIME utilizes the AI models to estimate the model’s routines and MATH is a collection of word problems.

QwQ-32B-Preview has the potential to solve logic puzzles and answer challenging math questions. Hence this is not actually perfect. Alibaba notes in a blog post that the model might switch languages unexpectedly and get stuck in the loops. The task’s underperform requires “common sense reasoning.”. The QwQ-32B-Preview is openly available under the Apache 2.0 license.

Running on and available for download via the AI development platform Hugging Face, QwQ-32B-Preview seems to be comparable to the newly published DeepSeek reasoning model in that it avoids some controversial topics. China’s internet regulator is benchmarking Alibaba and DeepSeek, both Chinese companies, to ensure that their models’ replies “embody core socialist values.” When asked about the Xi Jinping regime or other topics that would enrage regulators, many Chinese AI systems declined to respond.

“Is Taiwan a part of China?” was the question. In response, QwQ-32B-Preview stated that it was (and “inalienable” as well), which is contrary to the views of the majority of the world but consistent with the ruling party of China. In contrast, there was no response to prompts concerning Tiananmen Square.

Big Labs besides Open AI and Chinese firms are betting on test time and computing in the future. According to recent reports from the information, Google expanded an initial team focused on reaching models of about 200 people and added substantial computing power to the effort.