DeepSeek Releases Major Update to R1 AI Model
DeepSeek, the Chinese artificial intelligence startup, has unveiled the first update to its R1 reasoning model, as reported by Reuters. The update, named R1-0528, enhances the model’s depth of reasoning and inference capabilities, making it a stronger contender against global players like OpenAI and Google.
Although the update is labeled as a "minor version," DeepSeek claimed it significantly bolstered the model's ability to handle complex tasks, narrowing the performance gap with industry leaders like OpenAI’s O3 reasoning models and Google's Gemini 2.5 Pro.
This upgrade follows the initial launch of R1 in January, which went viral globally, challenging traditional beliefs around the need for massive computing power and investments to scale AI models. DeepSeek's R1 proved that more efficient models could rival the capabilities of established US tech giants.
Key Improvements and Features
According to Reuters, in a brief update on the X platform, DeepSeek revealed that R1-0528 offered a marked reduction in "hallucinations", false or misleading outputs, by approximately 45-50% in tasks such as summarizing and rewriting.
The company also highlighted the model's newfound capabilities in creative writing, including generating essays, novels, and other genres. Furthermore, the update has improved performance in generating front-end code and role-playing scenarios.
The improvements came after DeepSeek ran R1-0528 through various benchmark evaluations. The model has demonstrated excellent performance across multiple domains, including mathematics, programming, and general logic tasks, marking a significant leap in capabilities.
Also read: Microsoft Restricts Employees from Using DeepSeek Over Security and Propaganda Concerns
A Response to Rising Competition
The update comes amid increasing competition from Chinese tech giants like Alibaba and Tencent, who have released models claiming to surpass DeepSeek’s R1.
Despite this, DeepSeek continues to push forward, challenging the assumption that US export controls are stifling China's AI progress. The startup’s AI models have emerged as strong contenders in performance, with DeepSeek demonstrating that smaller models can deliver high performance at a fraction of the cost compared to their US counterparts.
DeepSeek utilized its R1-0528 model to refine and enhance Alibaba’s Qwen 3 8B Base model, resulting in a performance boost of over 10% and highlighting the value of its chain-of-thought reasoning, particularly for smaller models.
DeepSeek's R1-0528 update marks a key milestone as the company prepares for the delayed launch of R2, its next major model, prompting rivals like OpenAI and Google to lower prices and release lighter models in response.
