At Hyperbolic, we’re focused on delivering the latest open-source models as soon as they’re available to the developer community.
Today, we're the first to support DeepSeek-V3-0324, the newest update from the DeepSeek team on Hugging Face.
Released on March 24, 2025, DeepSeek-V3-0324 builds on the original DeepSeek-V3 (released December 2024) and features 685 billion parameters, up from 671 billion in the original. This includes an additional 14 billion parameters from the Multi-Token Prediction (MTP) module which is expected to boost performance, enabling faster speculative decoding during inference. The context window remains at 128,000 tokens, the same as DeepSeek-V3.
What is Hyperbolic’s Inference Service?
Hyperbolic’s Inference Service provides developers with fast, serverless deployment of open-source models. It removes the complexity of managing GPU infrastructure and offers one-click deployment, low-latency inference, and full privacy control with no data retention.
We support 25+ open-source models across text, image, vision-language, and audio, delivering inference at a fraction of the cost of major cloud providers while staying fully API-compatible with OpenAI and other ecosystems.
Key Benefits:
Instant API Access: Deploy models via REST API with support for Python, TypeScript, cURL, and Gradio—no infrastructure setup required. Compatible with OpenAI API format.
Scalable Inference Capacity: Handle high-demand AI workloads with flexible GPU availability.
Affordable Pricing: Pay-as-you-go pricing—no hidden fees or long-term commitments.
Custom Model Hosting: Deploy and run custom models directly on Hyperbolic’s infrastructure.
Low-Latency Global Infrastructure: Optimized for fast inference response times across regions.
Privacy-First Design: Zero data retention—no logging, tracking, or data sharing.
How do I access DeepSeek-V3-0324?
You can access DeepSeek-V3-0324 today on Hyperbolic’s Inference Service at this link.
Hyperbolic is also integrated with Hugging Face. You can access DeepSeek-V3-0324 by directly using Gradio within Hyperbolic or via the Hugging Face face playground using Hyperbolic's API here.
:format(webp))
We are committed to supporting the latest open-source models, and today’s integration is a reflection of that mission.
For the latest updates on DeepSeek, follow us on X.
About Hyperbolic
Hyperbolic is making global AI compute liquid. The platform turns fragmented GPU capacity into productive AI infrastructure, giving researchers, startups, developers, and AI-native companies one place to access on-demand GPUs, manage clusters, and run inference, without long waitlists, rigid contracts, or complex procurement cycles.
Founded by award-winning Math and AI researchers from UC Berkeley and the University of Washington, Hyperbolic is committed to creating a future where AI technology is universally accessible, verified, and collectively governed.
Website | X | LinkedIn | YouTube | GitHub | Documentation
:format(webp))
:format(webp))
:format(webp))
:format(webp))



:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))

:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))
:format(webp))