DeepSeek R1-0528 Now Available on Hyperbolic
Today we're announcing immediate availability of DeepSeek's latest R1-0528 model on the Hyperbolic platform. This update to the R1 reasoning model brings significant performance improvements while maintaining the open-source accessibility that made the original release so impactful.
Key Improvements
DeepSeek released R1-0528 early this morning with notable enhancements:
Hallucination rates reduced by 45-50% in rewriting and summarization tasks
AIME 2025 benchmark accuracy increased from 70% to 87.5%
LiveCodeBench accuracy improved from 63.5% to 73.3%
Enhanced capabilities for complex reasoning and code generation
The model maintains its 685B parameter architecture while delivering better inference quality through algorithmic optimizations.
API Immediate Availability
DeepSeek R1-0528 is now accessible through:
Hugging Face: huggingface.co/deepseek-ai/DeepSeek-R1-0528
We're serving the model in FP8 precision to ensure optimal performance for production workloads.
Infrastructure Requirements
Running R1-0528 effectively requires substantial compute resources:
For Inference:
Production: 10-20x NVIDIA H100 80GB GPUs
Alternative: 20x NVIDIA A100 80GB GPUs
Distributed setups with high-speed interconnects recommended
For Training/Fine-tuning:
Minimum: 256 nodes with 8x H100 GPUs each
Optimal: 2,000+ H100/H800 GPU cluster
Need GPUs to Train DeepSeek-R1-0528?
We offer flexible options for teams looking to deploy R1-0528:
On-Demand GPUs
H100s available at $0.99/hour
Instant provisioning for development and testing
No long-term commitments required
Business & Enterprise Cloud (Coming soon..)
Volume pricing for dedicated resources
99.9% uptime, 70% cost savings vs legacy cloud
No vendor lock in
Secure early bird pricing by connecting with our team HERE
Our infrastructure is designed specifically for large-scale AI workloads, with high-bandwidth networking and optimized storage systems.
Market Context
This release continues DeepSeek's pattern of delivering competitive performance at lower computational costs. The original R1 demonstrated that effective AI development doesn't require unlimited resources—the model was reportedly trained for $5.6 million compared to the hundreds of millions spent on comparable systems. Imagine if they would have trained their model with Hyperbolic? DeepSeek could have cut costs in half.
R1-0528's improvements, particularly in reducing hallucinations and enhancing reasoning capabilities, address key challenges in deploying LLMs for production use cases.
Getting Started
For teams ready to evaluate DeepSeek-R1-0528:
Access the model immediately at app.hyperbolic.xyz/models/deepseek-r1-0528
Review our API documentation for integration details at docs.hyperbolic.xyz
Contact our team for Business and Enterprise Cloud deployment options HERE
We're committed to providing reliable infrastructure for open-source AI models. As new models emerge, we'll continue expanding our platform to ensure developers have access to the latest innovations.
For questions about deploying DeepSeek R1-0528 or our GPU offerings, please reach out to our support team at [email protected].
About Hyperbolic
Hyperbolic is the on-demand AI cloud made for developers. We provide fast, affordable access to compute, inference, and AI services. Over 195,000 developers use Hyperbolic to train, fine-tune, and deploy models at scale.
Our platform has quickly become a favorite among AI researchers, including those like Andrej Karpathy. We collaborate with teams at Hugging Face, Vercel, Quora, Chatbot Arena, LMSYS, OpenRouter, Black Forest Labs, Stanford, Berkeley, and beyond.
Founded by AI researchers from UC Berkeley and the University of Washington, Hyperbolic is built for the next wave of AI innovation—open, accessible, and developer-first.
Website | X | Discord | LinkedIn | YouTube | GitHub | Documentation
