On May 28, 2025, Chinese AI startup DeepSeek released an updated version of its R1 reasoning model, named R1-0528, on the Hugging Face platform. Despite the lack of an official announcement, this update has garnered significant attention in the AI community for its performance enhancements and open-source accessibility.(Reuters)
Key points
- On May 28, 2025, DeepSeek released an updated version of its R1 reasoning model, named R1-0528, on the Hugging Face platform, without an official announcement.
- R1-0528 is fully open-source under the MIT License, allowing developers worldwide to access and use the model freely.
- The model has 671 billion parameters in total but activates only 37 billion during inference.
- Reuters reported that R1-0528 ranks just below OpenAI’s o4 mini and o3 models in code generation tasks, outperforming competitors such as xAI’s Grok 3 mini and Alibaba’s Qwen 3.
- Quantized versions reduce storage requirements from 700GB unquantized to approximately 33.8GB in a 1.78-bit dynamic format.
Key Features of DeepSeek R1-0528
- Open-Source Availability: R1-0528 is fully open-source under the MIT License, allowing developers worldwide to access and utilize the model freely.(Business Insider)
- Enhanced Reasoning Capabilities: The model demonstrates improved reasoning abilities, particularly in code generation tasks, where it ranks just below OpenAI’s o4 mini and o3 models, outperforming competitors like xAI’s Grok 3 mini and Alibaba’s Qwen 3 .(Reuters)
- Efficient Architecture: With a total of 671 billion parameters, R1-0528 activates only 37 billion during inference, optimizing performance without compromising computational efficiency .(Hacker News)
- Local Deployment Options: The model can be run locally, with quantized versions reducing storage requirements significantly—from 700GB unquantized to approximately 33.8GB in a 1.78-bit dynamic format (Unsloth Documentation).(Unsloth Documentation)
Performance Benchmarks
R1-0528 has been evaluated on various benchmarks, showcasing its capabilities:(Reddit)
- LiveCodeBench: Developed by UC Berkeley, MIT, and Cornell, this benchmark places R1-0528 just behind OpenAI’s o4 mini and o3 models in code generation tasks .(Reuters)
- Community Feedback: Early adopters have reported that R1-0528 successfully handles complex reasoning tasks, with some noting its ability to answer questions that other models, including Claude-4, could not .(Reddit)
Implications and Future Outlook
The release of R1-0528 underscores DeepSeek’s commitment to advancing AI capabilities while maintaining open-source principles. By offering a model that rivals top-tier competitors in performance and accessibility, DeepSeek continues to position itself as a significant player in the global AI landscape.(WIRED)
While the AI community awaits the anticipated release of DeepSeek’s R2 model, R1-0528 serves as a testament to the company’s ongoing innovation and impact on AI development.
Accessing DeepSeek R1-0528
Developers and researchers can explore and utilize the R1-0528 model through the following resources:
- Hugging Face Repository: DeepSeek R1-0528 on Hugging Face
- Local Deployment Guide: Running DeepSeek R1-0528 Locally
- API Access and Pricing: OpenRouter DeepSeek R1-0528
Leave a Reply