DeepSeek Upgrades R1 AI Model, Intensifies Competition With US Rivals
- tech360.tv
- May 30, 2025
- 2 min read
Chinese artificial intelligence start-up DeepSeek has released an upgraded version of its R1 reasoning model, intensifying competition with US tech giants such as OpenAI and Google.
The update, named R1-0528, was announced early Thursday via the developer platform Hugging Face. DeepSeek described it as a minor version upgrade that significantly enhances the model’s reasoning and inference capabilities, particularly in handling complex tasks.
The improved performance brings R1-0528 closer to OpenAI’s o3 reasoning models and Google’s Gemini 2.5 Pro.
DeepSeek’s original R1 model, launched in January, gained global attention and challenged the notion that scaling AI requires massive computing resources and investment. The launch also triggered a drop in tech shares outside China.
Unlike the R1 launch, which was accompanied by a detailed academic paper, Thursday’s update was initially released with limited technical information. However, DeepSeek later shared more details on social media platforms X and WeChat.
The company said R1-0528 reduces hallucinations—false or misleading outputs—by 45% to 50% in tasks such as rewriting and summarising. It also improves creative writing, front-end code generation, and role-playing capabilities.
DeepSeek stated that the model has shown strong performance in benchmark evaluations across mathematics, programming, and general logic.
The Hangzhou-based firm also revealed that it used the reasoning process of R1-0528 to enhance Alibaba’s Qwen 3 8B Base model through a method called distillation. The resulting model outperformed the original Qwen 3 by more than 10%.
DeepSeek said the chain-of-thought from R1-0528 could be valuable for academic research and industrial development of small-scale models.
The update comes amid growing competition in the AI sector. In response to DeepSeek’s rise, Google has introduced discounted access tiers for its Gemini model, while OpenAI has cut prices and launched the o3 Mini model, which uses less computing power.
DeepSeek is expected to release R2, the successor to R1. Reuters reported in March that the launch was initially planned for May. The company also released an upgrade to its V3 large language model in March.
DeepSeek released R1-0528, an upgraded version of its R1 AI model
The update improves reasoning, reduces hallucinations, and enhances creative tasks
R1-0528 brings performance closer to OpenAI’s o3 and Google’s Gemini 2.5 Pro
Source: REUTERS
