Lara Zijl 
Alibaba has released QwQ-32B, a new 32-billion parameter AI reasoning model that it claims outperforms larger competitors like DeepSeek’s R1 (671 billion parameters) and rivals OpenAI’s models in problem-solving tasks such as mathematics, coding, and general reasoning. Built upon Alibaba’s Qwen 2.5 language model, QwQ-32B processes text, images, and audio to analyze data, identify patterns, and generate solutions.
Alibaba highlights QwQ-32B’s efficiency, requiring significantly less data than competing models. The model incorporates reinforcement learning (RL) and agent capabilities, enabling critical thinking, tool utilization, and adaptive reasoning based on environmental feedback. Alibaba believes these advancements bring them closer to achieving Artificial General Intelligence (AGI) by combining strong foundation models with scaled RL resources. The company is actively exploring the integration of agents with RL for enhanced long-horizon reasoning and faster inference times.
