DeepSeek-R1-0528 Update Narrows Gap with OpenAI's Top Coding Model

DeepSeek, a Chinese AI startup, has released an upgraded version of its large language model—DeepSeek-R1-0528—on the open-source platform Hugging Face. Although official documentation has yet to be updated, developer benchmarks on LiveCodeBench indicate that the new model's coding performance nearly matches OpenAI's o3, the current leader in reasoning capabilities.

The update was quietly rolled out late yesterday as a “minor version trial”, according to a notice in DeepSeek's official community group. However, developers emphasized that the improvements are far from minor.

In addition to stronger coding performance, DeepSeek-R1-0528 also delivers enhancements in writing fluency, formatting, and natural language generation. Its standing relative to Anthropic's Claude Opus 4—the current benchmark for coding models—remains uncertain.

Back in March, DeepSeek released DeepSeek-V3-0324, incorporating reinforcement learning from R1's training. That update significantly improved performance in reasoning, front-end development, and Chinese writing.

While the long-awaited DeepSeek-R2 has yet to be announced, the strong performance of the latest R1 upgrade has raised user expectations. Developers are hoping R2 will bring extended context length and multimodal capabilities—key features for production-level deployment.