Chinese AI startup DeepSeek released an update to its R1 reasoning model on Thursday, intensifying competition with U.S. companies like OpenAI and Google.
The updated version, R1-0528, improves reasoning depth and inference capabilities, enhancing performance on complex tasks and narrowing the gap with OpenAI’s o3 models and Google’s Gemini 2.5 Pro.
Since its January launch, R1 gained global attention, impacting tech shares outside China and challenging assumptions about the resources required for AI development. In response, Chinese tech giants Alibaba and Tencent have introduced models that claim to outperform DeepSeek’s.
DeepSeek stated the update reduces false or misleading outputs by nearly half in tasks such as rewriting and summarizing. The model now supports creative writing, code generation, and role-playing, while demonstrating strong results in math, programming, and logic benchmarks.
The company also applied the reasoning process from R1-0528 to enhance Alibaba’s Qwen 3 8B Base model, boosting its performance by over 10%. DeepSeek emphasized the importance of this advancement for academic research and industrial development of smaller AI models.
The update follows efforts by competitors like Google and OpenAI to offer lower-cost, more accessible AI solutions. DeepSeek is expected to release a successor, R2, with earlier reports indicating a possible launch in May. The company also upgraded its V3 large language model in March.
