In the race for Artificial Intelligence, it seemed that companies such as Google, Microsoft and OpenAI would be in the lead for years. Fortunately, the reality is different and the AI proposals coming out of China have managed to capture the attention of a large number of users.
The reason for this is that their capabilities leave the leading AI models in the dust, none of them escape this competition, and DeepSeek is certainly no exception.
Qwen has just proved once again that it has no rivals with the announcement of QwQ-32B as a new model that is capable of putting DeepSeek-R1 in its place.
Qwen proves once again that AIs coming from China can change the paradigm.
To understand what the launch of Qwen’s QwQ-32B model means, we have to go back to January of this year. Just two months ago DeepSeek surprised the world with an AI model capable of putting OpenAI’s proposals in their place. We didn’t have to wait long for DeepSeek to find competition; Qwen came along to take the AI down a peg or two.
While the initial launch of Qwen put DeepSeek-V3 in check, now the new Qwen model is putting DeepSeek-R1 in check. What we have been able to learn from the release of this model is that a QwQ-32B outperforms DeepSeek-R1 in several areas. But the most surprising thing about this is that it does so in a model that is twenty times smaller than DeepSeek’s proposal.
This smaller model is open source, like the previous version, has 32 billion parameters for complex reasoning, which sounds small compared to DeepSeek’s 671 billion. But this is what DeepSeek-R1 is able to cope with, surpassing it in some tests and situations.
Specifically, QwQ-32B outperforms DeepSeekR1 in tests such as LiveBench, IFEval and BFCL. However, it is close in other areas such as AIME24 and LiveCodeBench. In addition, Qwen’s new model integrates reinforcement learning, which allows the model to improve its reasoning by learning from trial and error.
How to test QwQ-32B the new Qwen model?
The new model is open source, all users can access it thanks to this philosophy and integrate it into their projects or make their own projects.
Qwen makes available on its official blog all the necessary information to access this new model.
