DeepSeek, a Chinese artificial intelligence startup, has announced an update to its R1 reasoning model.

This development reportedly intensifies the competition within the AI sector, particularly with US-based firms such as OpenAI and Google.

The R1-0528 update is said to significantly improve reasoning and inference capabilities, narrowing the gap with OpenAI’s o3 models and Google’s Gemini 2.5 Pro, reported Reuters.

The update to the R1 model, described as a “minor trial upgrade,” was announced on a WeChat group and reported by Bloomberg.

Users have been invited to commence testing the new version.

According to the details shared on Hugging Face, a developer platform, the R1-0528 update enhances complex task management.

The initial launch of the R1 model in January went viral globally and negatively affecting technology stocks outside China. It also challenged the notion that successful AI scaling requires substantial computing resources and investment. Since then, Chinese tech conglomerates such as Alibaba and Tencent have released their own models, claiming advancements over DeepSeek’s offerings.

Despite the significant attention the R1 model garnered at its launch, the latest update was released with fewer details. However; DeepSeek later disclosed on X that the R1-0528 version boasted improved performance. On WeChat, the company reported a 45-50% reduction in “hallucinations,” the AI’s false or misleading outputs, particularly in tasks such as rewriting and summarising.

DeepSeek also highlighted the update’s new abilities, including creative writing for essays and novels, enhanced front-end code generation, and role-playing capabilities.

“We believe that the chain-of-thought from DeepSeek-R1-0528 will hold significant importance for both academic research on reasoning models and industrial development focused on small-scale models,” the company stated.

In the wake of DeepSeek’s advancements, competitors such as Google’s Gemini have introduced discounted access levels, and OpenAI has reduced prices, releasing an o3 Mini model that requires less computational power. Despite these updates, there is widespread anticipation for the release of R2, DeepSeek’s successor to the R1 model.

In March 2025, Reuters reported that R2 was scheduled for a May release. DeepSeek also updated its V3 large language model in March.