DeepSeek’s R1-0528 AI Model Rivals OpenAI in Code Generation

Chinese AI startup DeepSeek has quietly released an upgrade to its R1 reasoning model, intensifying competition with U.S. leaders like OpenAI. The new model, R1-0528, is now available on Hugging Face, although the company has not yet issued an official announcement or technical documentation.

Despite the low-key release, the R1-0528 model quickly drew attention after climbing the LiveCodeBench leaderboard—a benchmark co-developed by researchers at UC Berkeley, MIT, and Cornell. The model ranked just behind OpenAI’s o4 mini and o3 models in code generation, while outperforming xAI’s Grok 3 mini and Alibaba’s Qwen 3.

According to Bloomberg, a DeepSeek representative described the release as a “minor trial upgrade” in a WeChat developer group, inviting early users to begin testing.

Earlier this year, DeepSeek made headlines with the launch of its original R1 model, challenging assumptions that U.S. export restrictions were slowing China’s AI progress. R1 shocked the global tech market by matching or exceeding leading U.S. models in performance—at a fraction of the cost. Its release in January triggered a drop in tech shares outside of China and reshaped industry perspectives on the role of scale and computing power in AI development.

Since then, Chinese tech giants like Tencent and Alibaba have released competing models. Meanwhile, OpenAI has responded by launching its o3 Mini, optimized for lower compute usage, while Google’s Gemini has introduced more affordable access tiers.

DeepSeek is also expected to release its R2 model, a successor to R1, with sources earlier indicating a May timeline. In March, the company released an upgrade to its V3 large language model, further expanding its AI portfolio.