Alibaba has unveiled a new AI model called QwQ-32B-Preview, at a time of intense competition among tech giants trying to compete with each other with cutting-edge and potent AI tools.
The Qwen team of the Chinese e-commerce giant reportedly built the tool to compete with OpenAI’s ChatGPT o1, the first AI model to be freely available for download under a permissive license. With 32.5 billion parameters, the AI model can understand prompts as long as 32,000 words, which is its most noteworthy feature.
Alibaba’s Latest AI Model
In comparison to ChatGPT o1-preview and o1-mini, QwQ-32B-Preview has shown to be an extremely strong rival to OpenAI’s two reasoning models, outperforming them on specific benchmarks.
In artificial intelligence, parameters represent a model’s ability to solve problems; hence, models with more parameters tend to outperform those with less.
Alibaba’s AI Model Outperforms OpenAI on Key AI Tests
The benchmark threshold of the QwQ-32B-Preview demonstrates that OpenAI’s reasoning models have fewer parameters than Alibaba’s AI model, given that OpenAI has not disclosed the parameter count for its reasoning models thus far.
Internal Alibaba studies show that QwQ-32B-Preview outperforms OpenAI’s o1 models on the AIME and MATH tests, respectively. The AIME test utilizes third-party models to evaluate a model’s performance, while the MATH test employs word problems as a whole.
The “reasoning” skills of Alibaba’s QwQ-32B-Preview can solve logic puzzles and math problems that aren’t difficult enough.
However, there are a few drawbacks to consider. According to a blog post on the official Alibaba website, it can switch languages abruptly, become trapped in loops, and misunderstand cues that need “common sense reasoning.”
