China's DeepSeek Unveils Update to Its R1 Reasoning Model

DeepSeek, a Chinese artificial intelligence startup, released an update to its R1 reasoning model early Thursday, stepping up competition with US rivals such as OpenAI.

Key Highlights

• DeepSeek's R1-0528 update ranks just below OpenAI's o3, surpassing xAI's Grok 3 mini and Alibaba's Qwen 3 on LiveCodeBench.
• The R1-0528 update, released on Hugging Face without formal announcement, intensifies competition with U.S. AI leaders.

DeepSeek released R1-0528 on the developer platform Hugging Face but has yet to make an official public announcement. It did not publish a model description or comparisons.

However, the LiveCodeBench leaderboard, developed by researchers from UC Berkeley, MIT, and Cornell, placed DeepSeek's updated R1 reasoning model just behind OpenAI's o4 mini and o3 reasoning models on code generation, but ahead of xAI's Grok 3 mini and Alibaba's Qwen 3.

Bloomberg previously reported the update on Wednesday. It stated that a DeepSeek representative informed a WeChat group that it had completed a "minor trial upgrade" and that users could begin testing it.

DeepSeek debunked the notion that US export controls were impeding China's AI advancements earlier this year, when the startup released AI models that were on par or better than industry-leading models in the United States at a fraction of the cost.

Also Read: South Korea Accuses DeepSeek of Transferring User Data to China, US Without Consent

The launch of R1 in January sent tech shares outside China plummeting and called into question the notion that scaling AI requires massive computing power and investment. Since R1's release, Chinese tech titans such as Alibaba and Tencent have released models that claim to outperform DeepSeek's.

Google's Gemini has introduced discounted tiers of access, while OpenAI has reduced prices and released an o3 Mini model that uses less computing power.

The company is still widely expected to release R2, a follow-up to R1. According to sources, R2 was originally scheduled to be released in May. DeepSeek also released an update to its V3 large language model in March.