Alibaba’s Qwen Team Releases QwQ-32B Open-Source Reasoning Model, Said to Perform Similar to DeepSeek-R1

Alibaba’s Qwen Team, a division tasked with creating synthetic intelligence (AI) fashions, launched the QwQ-32B AI mannequin on Wednesday. It is a reasoning mannequin based mostly on prolonged take a look at time compute with seen chain-of-thought (CoT). The builders declare that regardless of being smaller in dimension in comparison with the DeepSeek-R1, the mannequin can match its efficiency based mostly on benchmark scores. Like different AI fashions launched by the Qwen Team, the QwQ-32B can be an open-source AI mannequin, nevertheless, it isn’t totally open-sourced.

QwQ-32B Reasoning AI Model Released

In a blog post, Alibaba’s Qwen Team detailed the QwQ-32B reasoning mannequin. QwQ (brief for Qwen with Questions) sequence AI fashions had been first launched by the corporate in November 2024. These reasoning fashions had been designed to supply an open-source different for the likes of OpenAI’s o1 sequence. The QwQ-32B is a 32 billion parameter mannequin developed by scaling reinforcement studying (RL) methods.

Explaining the coaching course of, the builders stated that the RL scaling strategy was added to a cold-start checkpoint. Initially, RL was used just for coding and mathematics-related duties, and the responses had been verified to make sure accuracy. Later the method was used for normal capabilities together with rule-based verifiers. The Qwen Team discovered that this methodology elevated normal capabilities of the mannequin with out lowering its math and coding efficiency.

QwQ-32B AI Model benchmarks
Photo Credit: Alibaba

 

The builders declare that these coaching constructions enabled the QwQ-32B to carry out at comparable ranges to the DeepSeek-R1 regardless of the latter being a 671-billion-parameter mannequin (with 37 billion activated). Based on inside testing, the crew claimed that QwQ-32B outperforms DeepSeek-R1 within the LiveBench (coding), IFEval (chat or instruction fine-tuned language), and the Berkeley Function Calling Leaderboard V3 or BFCL (potential to name capabilities) benchmarks.

Developers and AI fans can discover the open weights of the mannequin on Hugging Face itemizing and Modelscope. The mannequin is accessible beneath the Apache 2.0 licence which permits educational and research-related utilization however forbids industrial use circumstances. Additionally, because the full coaching particulars and datasets should not accessible, the mannequin can be not replicable or might be deconstructed. DeepSeek-R1 was additionally accessible beneath the identical licence.

In case one lacks the suitable {hardware} to run the AI mannequin regionally, they’ll additionally entry its capabilities by way of Qwen Chat. The mannequin picker menu on the top-left of the web page will let customers choose the QwQ-32B-preview mannequin.