DeepSeek R1 Distill Qwen 32B Compare

DeepSeek R1 Distill Qwen 32B

DeepSeek · → text

The model provider is the Sophnet platform. Deepseek-R1-Distill-Qwen-32B is a knowledge-distilled large language model based on Qwen 2.5 32B and trained using outputs from DeepSeek R1. DeepSeek-R1 addresses issues such as infinite repetition, poor readability, and language mixing by introducing cold-start data before reinforcement learning. DeepSeek-R1’s performance in mathematics, programming, and reasoning tasks is comparable to OpenAI-o1. To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini on multiple benchmark tests, setting new state-of-the-art results for dense models.