Hyperbolic GPU Cloud & Inference

Offering overview

Capacity on Hyperbolic comes from a network of partner data centers unified behind one account, which lets the company list on-demand H100, H200 and B200 instances with no commitment and refresh prices weekly against supplier rates. Reserved clusters with sub-one-year commitments and a Private Cloud tier with isolation and compliance controls cover longer-term needs, while a product called Forge provides the provisioning layer to compute providers themselves. On the software side, a serverless inference API serves open-weight text, vision and image models (Llama 3.1 8B through 405B, Qwen2.5 72B and Coder 32B, DeepSeek, Pixtral, Qwen2-VL, Stable Diffusion and FLUX), and dedicated model hosting is available for custom checkpoints. Pricing is usage-based: H100 SXM bare metal and VMs were announced from $1.49 per GPU-hour, Llama 3.1 8B base inference lists at $0.10 per million tokens, and image generation starts around $0.01 per 1024x1024 image. Startups, research labs and inference platforms that want cheap, interruptible-free Hopper capacity without long contracts are the core audience; customers named on the site include AI labs and universities.

Technical specifications

Containers

Unknown

Kubernetes

Unknown

Root access

Unknown

API access

Unknown

Autoscaling

Unknown

Persistent processes

Unknown

AI capabilities

Model serving

Llama 3.1 8B, Llama 3.1 70B, Llama 3.1 405B, Llama 3.2 3B, Qwen2.5 72B, Qwen2.5 Coder 32B, Qwen2-VL, DeepSeek-V2.5, Hermes 3, Pixtral 12B, Stable Diffusion, FLUX

AI frameworks

Not stated

Inference endpoints

Unknown

Fine-tuning

Unknown

Vector database

Unknown

Model catalog

Unknown

Monitoring

Not stated

Evaluation

Not stated

Security & compliance

Geography

Provider headquarters are separate from offering infrastructure and availability.

Available plans

Pricing can change. Promotional prices are shown separately from standard prices.