DeepSeek V3.1 surpasses OpenAI GPT-OSS-20B in reasoning and programming.

DeepSeek has just issued the V3.1 model with a hybrid architecture, drawing attention for outperforming OpenAI's GPT-OSS-20B in most tests. Results show that DeepSeek generates complete code and works on the first attempt, while GPT-OSS-20B continuously fails, even freezing in high inference mode.

In the realm of creative writing, DeepSeek also excels with its logical, detail-rich storytelling, while OpenAI leans towards philosophy but lacks coherence. Additionally, DeepSeek effectively tackles complex reasoning tests, while GPT-OSS-20B often falls into meaningless loops.

However, OpenAI still holds an advantage in customization capabilities with many specialized versions rapidly developed by the community, while DeepSeek is just beginning to be refined.