DeepSeek-R1-0528 Release Notes

DeepSeek has released DeepSeek-R1-0528, an updated version of the R1 model designed to improve reasoning performance and developer utility. This release focuses on enhancing the model's accuracy, reducing hallucinations, and expanding its capabilities for structured data output and tool integration.

Key Model Improvements

DeepSeek-R1-0528 demonstrates improvements across several core areas of model performance:

  • Benchmark Performance: The model features improved performance on standardized benchmarks compared to previous iterations.
  • Hallucination Reduction: DeepSeek has implemented enhancements to reduce the frequency of hallucinations, increasing the reliability of the model's generated responses.
  • Front-end Capabilities: The release includes enhanced capabilities for front-end interactions.

Developer Features and Integration

DeepSeek-R1-0528 introduces specific features to make the model more suitable for programmatic use and agentic workflows:

  • JSON Output Support: The model now supports JSON output, allowing for structured data extraction and more reliable integration into software pipelines.
  • Function Calling: The model supports function calling, enabling it to interact with external tools and APIs to perform specific tasks.

Deployment and Access

DeepSeek-R1-0528 is available through multiple channels:

  • Chat Interface: Users can access the model via chat.deepseek.com.
  • API Access: The model is available via the DeepSeek API. There are no changes to API usage requirements for this update.
  • Open-Source Weights: The model weights have been made available on Hugging Face at huggingface.co/deepseek-ai/DeepSeek-R1-0528.

Technical Documentation

For developers implementing the model, DeepSeek provides specific guidance on "Thinking Mode," which is a core component of the R1 series' reasoning process, detailed in the official Thinking Mode guide.

Sources

Related

  • Dispatch
    DeepSeek-V3-0324 Release Notes

    DeepSeek has released DeepSeek-V3-0324, an updated version of the base model for R1 that features improved instruction following, coding, and math capabilities under an MIT license.

  • Dispatch
    DeepSeek-V3-0324 Release Notes

    DeepSeek has released DeepSeek-V3-0324, featuring improved reasoning, front-end development, and tool-use capabilities, now available under the MIT License.

  • Dispatch
    DeepSeek-V3.1 Release Notes

    DeepSeek-V3.1 introduces hybrid inference modes, improved agentic capabilities, and continued pretraining for long context extension, marking a step toward agent-centric AI.

  • Dispatch
    DeepSeek-V3.2 and DeepSeek-V3.2-Speciale Release

    DeepSeek has released DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, reasoning-first models designed for agents that introduce integrated thinking in tool-use and gold-medal performance in competitive programming and mathematics.

  • Dispatch
    Open R1 Update #2: OpenR1-Math-220k Dataset and Reasoning Insights

    Hugging Face introduces OpenR1-Math-220k, a large-scale math reasoning dataset designed to reconstruct DeepSeek R1's distillation pipeline, while sharing community insights on GRPO and Chain-of-Thought length control.